domejunky

KWorld UB499-2T under Arch Linux

2012-01-13T00:56:00.000Z

I've just been wresting with this so I thought I'd save someone some time:

This is the usb-dvb-t device mentioned on this site - I've blogged about using it under Ubuntu before but things have changed for 3.x kernels, and I'm using Arch more and more now. This might all be redundant when 3.2 kernels arrive - since the driver is in the kernel tree.

I wasted a whole evening following the instructions linked from the above site to get the v4l drivers going. No joy under Arch - the process worked but I got all sorts of 'disagree with symbol' errors, and it hosed the drivers for my dvb-s device. I had to do a 'make rminstall' in the v4l source - and then 'pacman -S linux' to get my original Arch drivers back (I'd deleted them since Arch uses compressed modules, and that was sited as one of the potential sources of the errors). I also did a 'pacman -S v4l-utils' - but I don't think I actually needed to...

So I went back to the non-v4l driver. I downloaded this (for 64bit in my case) This is meant only to support kernels up to 2.6.39 - but adding:

ifneq (,$(findstring 3.1.8,$(CURRENT)))
@cp -f v4l/kernel-3.1.8/* ./
endif

...to the Makefile, and copying v4l/kernel-2.6.39 to v4l/kernel-3.1.8 I managed to get it to compile. I failed at the this installation, but I copied the 'dvb-usb-it9135.ko' to '/lib/modules/3.1.8-1-ARCH/kernel/drivers/media/dvb/dvb-usb' modprobed it, and Kaffeine picked it up without even restarting it...cool!

Django + Nginx and shared media

2011-12-02T00:48:00.001Z

I've been struggling with this for a couple of days - so I'm sticking them here for memoryade. I'm building a site from Django, using Nginx as the front end.

Maybe I'm suffering from Apache hangovers, but I've been having problems getting a nice solution to URLs mapped to views based off the same template. I think things were confused by also wanting to map the root of the site to a Django URL of '/'. Probably bad form, but I basically used my 'index.html' as a template, with Django style {%...%} blocks to be overridden by templates for each section. This means I have a Django project with no apps defined at the beginning, save for admin - which has it's own media mapping in Nginx.

The problem was that I managed to configure Nginx to map to '/' - and I could get Django to respond with different views in the root app - things like 'index' and 'projects' - but I would loose all the media. They would always append their part of the URL when looking for media like images and style-sheets. So if I went to 'projects' all links to 'media/image.png' would end up as 'projects/media/image.png' - I think Django expects all apps to contain their own media directory. So partly through thinking that I was messing with things having defined views in the default Django app, and mapped it to '/' - spent 2 days trying all sorts of Nginx and 'urls.py' configurations.

The answer was stupid - edit the original template to use links mapped to '/' so instead of "media/image.png" put "/media/image.png" - relative to the root, rather than the section your in.

So this is what I ended up with:

urls.py

urlpatterns = patterns('',
('^$', index),
('^projects/$', projects),
# Uncomment the next line to enable the admin:
url(r'^admin/', include(admin.site.urls)),
)

views.py

def index(request):
return render_to_response('home.html', {'content': "HELLO THERE!"})
def projects(request):
return render_to_response('home.html', {'content': "Projects"})

I'm using home.html twice, just adjusting the content with different views. 'home.html' looks like this

{% extends "index.html" %}
{%block content%} {{content}} {%endblock%}

Then the Nginx configuration just looks like this:

server {
#listen 80; ## listen for ipv4; this line is default and implied
#listen [::]:80 default ipv6only=on; ## listen for ipv6
root /home/me/Projects/mysite;
#index index.html index.htm;
# Make site accessible from http://localhost/
server_name localhost;
location /media {
alias /home/me/Projects/mysite/media ;
}
location /admin/media {
alias /home/me/Projects/mysite/mysite/admin/media;
}
location / {
# First attempt to serve request as file, then
# as directory, then fall back to index.html
#try_files $uri $uri/ /index.html;
fastcgi_pass 127.0.0.1:8000;
fastcgi_param PATH_INFO $fastcgi_script_name;
fastcgi_param REQUEST_METHOD $request_method;
fastcgi_param QUERY_STRING $query_string;
fastcgi_param CONTENT_TYPE $content_type;
fastcgi_param CONTENT_LENGTH $content_length;
fastcgi_pass_header Authorization;
fastcgi_intercept_errors off;
fastcgi_param SERVER_NAME $server_name;
fastcgi_param SERVER_PORT 80;
fastcgi_param SERVER_PROTOCOL wsgi;
}
}

I added the last three myself - I hadn't seen them used anywhere - but it stopped my logs being full of error about them missing....

This approach works perfectly, and can be moved around, chopped up onto different servers etc...

Arch...

2011-10-20T19:17:00.002+01:00

I've been playing with Arch Linux recently. Actually I've been playing with Debian, OpenSuSE, Xubuntu plus a few others - ever since Ubuntu 11.10 took away a workable Gnome fallback. Looks like I'm not alone, and I knew it was coming. Unity and Gnome Shell are garbage...

I've had a good relationship with Ubuntu since about 2006. I've used Linux on a lot of laptops, from a G3 Pismo in 2001, various HPs, Dells, Samsungs and a first generation Intel Macbook Pro - with varying amounts of grief. Ubuntu lowered the grief level, and had a polished interface. I also build IGs for immersive environments, and have workstations all over the place - it was good to not have to spend hours configuring a Linux desktop to get a workable and consistent environment. On servers I've used FreeBSD, RedHat/CentOS, SuSE and Debian mostly, so I know my way around different package managers, config files etc - I suppose I had most affinity with Debian, so Ubuntu fitted. Like a polished Debian. I have a lot of respect for FreeBSD, it's just such a well thought out system, but as my development stack got more precarious, it became difficult to use it for graphical and audio duties.

Every now and then I'd install Xubuntu over Ubuntu - and I quite liked it. It was quick, but often hard on the eye when you've become accustomed to Ubuntu. With the Gnome/Unity debacle, I decided (like others) to adopt it permanently. With the "Greybird" theme - it actually looked quite slick. So I stuck Xubuntu on a few boxes, and starting getting back on with things. But, an itch started forming. I figured that if I'd weened myself off the eyecandy, I might as well jump ship to another platform, and benefit from a more conservative attitude to updates and software choices. You can go and read a few reviews of distributions all over the net, but often they're written by people who've only installed them into a virtual machine - with analysis exngtending about as far as the colour of the desktop. To really test a distribution you need to install it onto real hardware - and the hardest test I have is my first generation Macbook Pro. Old school ATI GPU and EFI make for a rough ride. Initially I thought I'd go back to Debian - but EFI and Debian isn't fun, and the documentation is all over the place. Mint was the same. OpenSuSE works like a charm with EFI, and the original SuSE was the first distribution I ever got friendly with. The problem is I've never been totally happy with RPM. There's nothing wrong with it, just unfamiliarity, and a few of the tools I was after (Panda3D and CLAM-audio) would need compiling from source.

While I was fishing around I gave Arch a try. It worked with EFI for a start - which was quite a shock given its paired down nature. More and more though I'm liking it. It feels quite FreeBSD-ish, especially with 'rc.conf' and no SysV - but all the weird bits of software I need are there in binary form. There's quite a bit of work to configure things, but the mystery of all the stuff under the hood has disappeared - not the low level things, but the bit's that had made Ubuntu so comfy, automounting usb drives, wlan, laptop lid and suspend...that kind of thing. Arch has amazingly clear and succinct documentation, and really sane config files - so you find yourself being able to configure things really quickly. I even found a way to get a '#' out of my Mac keyboard, I even added '°' for good measure. Just for reference here's the magic line in ~/.Xmodmap

keycode 49 = numbersign degree

I tried to setup a Nginx + Django server on Amazon's EC2. It only took me 25 minutes - I was amazed! The only rub could be the rolling release schedule - I'm looking for a way to pin an install to a particular point in time so that I can be confident that my dev box will be in sync with deployment boxes...weĺl see. Other than that, I think I've found a new home....

Arduino Dome Control

2011-09-15T09:34:00.000+01:00

I'm looking for a way to replace the AMX/Netlinx in the dome. This is as proprietary as it gets - AMX panel is expensive, the Netlinx is expensive, the SDK is expensive - and the whole shebang ends up completely non extensible without access to a £400 per day contractor.
The Netlinx is basically a bunch of RS232 (and InfraRed) connections that talk to the Lights, Signage, Projector and Extron Matrix etc.

It occurred to me that the AMX panel is basically the same as the iOS/Android/~~WebOS~~ tablets that are available at the moment - and that I could do RS232 with some arduinos. I have some Python code running on the boxes in the dome, which uses sockets to send commands. This has been working very successfully to start shows and for emergency shutdown - my plan was to modify this to send RS232 and InfraRed. I'm working from home at the moment, so the test bed was an LG LCD TV (which has RS232 on the back) and my Onkyo home cinema amp (via IR).

I started writing the app in the emulator that came with the Android SDK, but this seemed to have a problem with UDP sockets. It emulates a 3G connection, but not WiFi. So I got a shitty 7" Tablet from ebay for £26. I managed to write a small Android app with a few buttons that would trigger certain RS232/IR signals.

I dug my USB-RS232 adpater out. After a lot of pouring over manuals for RS232 codes I managed to get this working. You can test the strings they give you in the manual in a serial terminal - but there is no standard notation in the manuals, you never know if your sending ASCII, Hex etc. The python serial library works a treat though. There is an IR library for Arduino which encapsulates a lot of the codes for common devices - this didn't work for my 1st Gen Arduino, so off to ebay again, and £12 for an Arduino Nano...

It all works like a champ - Java is still a bad language, and I need to find a way to enumerate multiple USB-RS232 devices to make sure I know which one is plugged into which device...

Abdij

2011-05-02T19:30:00.000+01:00

I've started brewing my own beer. I'm trying to brew a 'Leffe Bruin' clone. This seems quite a difficult style to start with, so I might call it 'Hair Shirt'...

My interest was piqued when I was staying in a friend's house - one of the other house-mates was brewing Cider. I have an aversion to apples so this was quite difficult - but I did taste a maple syrup version and it was good.

I'd also been drinking a lot of beer recently, not excessive amounts, but I wide variety. Mostly Belgian, which doesn't come cheap. It's Sainsburys' fault really - they had a 2 for £5 on the 750Ml bottles of Leffe Blond and Bruin - they really help shorten the A303. Then the other supermarkets sort of followed suite. Moving town, and not having a regular supermarket, has led me to stray off the beaten track looking for beers as interesting and as cheap as the Leffe deal...

Lidl do a remarkable little lager called Finkbräu - this says it's brewed to the reinheitsgebot, it's £3.80 or something for 6. Still not as interesting as a Leffe. I wandered through the Bitters, the odd Hoegaarden, I'm partial to Samuel Adams or Blue Moon which I discovered in the US, but they are quite pricey - and Samuel Adams disappeared completely for about 9 months...

So I thought I'd have a go myself. I found a brewing supplies place near me, bought a barrel etc I bought an extract kit from Brewferm - a Belgian kit supplier - it's been 4 days now and she's been bubbling like a good'n...nothing really innovative yet, I'm just making sure I understand the process.

Along the way I discovered this guy on youtube - fantastic viewing, I'm gonna send him some beer:

Mike's Ultrarad Beer Reviews

Fisheyes in the Clouds

2011-04-06T23:09:00.005+01:00

I'm rendering a piece for a competition at this year's Jena Domefest. The piece comprises various visual data representations of a work by Liszt. This needs to rendered at 2k fisheye - that is 2048x2048. It should be rendered at 3.2k - but I'm entering as a student.

I've constructed the piece using two different audio -> IPO curve scripts in Blender 2.49 - One called 'Audio Analysis', the other 'Sound tracker'. I then opened the Blender 2.49 files in Blender 2.56 and re-saved - I could then import the IPO curves, now 'f-curves', into the current file. The f-curves are set up to manipulate camera position, lamp power - but with Blender's new 'everything is animatable' attitude, I could have set key frames for anything...The main object in the scene is a volumetric rendering of a spectograph I generated with Arss.

I'm using Yafaray as the render engine as it has a native fisheye camera. The Yafaray exporter for Blender 2.56 is coming along nicely. When it works, due in-part to Blender's new plugin APIs, you wouldn't know that Yafaray wasn't a native part of Blender. All material panels change to show Yafaray options rather than Blender.

While I was working I rendered at 25% - I did test renders from time-to-time at full resolution - to make sure I still had time to do the final render. I rigged together an 8 core and two 4 core machines, all running Blender from, and rendering to, an NFS share - works pretty well. When it came time to render out at 2k somehow my average render time was something like 4 minutes per frame...Ouch! I would need 21 days to render, and a quick glance at my Gantt chart showed only 5...

I started thinking about who I could borrow machines from...we have a Dell Blade chassis in Plymouth, with two 8 core blades, and a couple of 8 core Mac Pros - but that's the other side of the country. So I started thinking about the Cloud...The Blade Chassis had given me enough exposure to virtual machines, so I setup an Amazon AWS account, and 10 minutes later I had a Ubuntu box up and running. I transfered my project files and a copy of Blender and Yafaray to the machine - I needed to install Pythonlibs and also transfer my config files from ~/.blender/2.56 - and off we go....rendering in the cloud. Except we didn't. The EC2 machines are really slow - either network latency, but more a case of actually stopping every 20 secs for 20-30 secs. The end result was a render time of 17 minutes, or rather 17 minutes until Blender/Yafaray were killed. Just for comparison, I looked around for another free Cloud trial. I found a company called ElasticHosts. They have an amazingly simple interface, and you end up with a much faster machine, that feels like a proper machine. Also no private/public keys - just straight ssh. I got my setup onto the box, and off we went...and this time it was quite spritely. The image above was rendered entirely on the Elastic box. Elastic reckon this is equivalent to a 2GHz single core Opteron - which seems about right. I got a render time of 231 secs, which is about 5 times slower than the 8 core Xeon based machine I have.

While working out how to kick off Blender from the command line with Yafaray, I found that render-times are radically different between Blender GUI and command line - maybe by a factor of 10 on some frames. So my problem was solved - my render times dropped to well under a minute....

memoryade:

Ubuntu 10.10
#tasksel -> install XFCE Live CD
Transfer Blender + Yafaray
[Transfer .deb for Yafaray]
Transfer .blend files and textures
Transfer ~/.blender/2.56/config to same place on server
Install Pythonlibs for 3.1
Add Universe and Multiverse repos
Install Blender 2.49 from repos

Métamorphose

2011-03-07T19:30:00.002Z

Let's hear it for Metamophose a bulk file renamer. I used to swear by Thunar - part of XFCE. But I've just had to rename 4000 TGA files, out of 14,000 TGA files to have proper zero padding and Thunar gave up. Metamophose was the only thing that could handle it...

KWorld UB499-2T vs Windows

2011-02-22T00:02:00.007Z

This is a total rant, I'm not even going to apologise.

I bought my Girlfriend a USB FreeView (DVB-T) device for Christmas. I was a cheap one from Maplin, and it was half price - a KWorld UB499-2T. I thought this was mainstream enough to have Mac OS X support, so that she could use it with her laptop when she was away from home. It didn't. No Worries, we used it in our bedroom with a laptop rebooted into Windows. The software that came with it was ugly, really poor HCI, but it worked. Since Christmas, day by day, the performance has become worse and worse. Each day the video got choppier, more artefacts appeared, skipped keyframes etc. I put this down to atmospherics since we've had quite a lot of foggy weather, snow etc...

I was aware of a driver for Linux, but it needed the module compiling, and laziness stopped me getting round to installing it. Tonight I gave in, the video under Windows was unwatchable. So I followed the instructions here

Pretty painless really. Modprobed the driver, set Kaffeine to scan, and up popped the channels, plus a couple of extra ones...shweet...

I use Windows and Mac OS X as little as possible these days, and you tend to forget just how bad they are. I've been using Windows since 3.11 and supporting Windows since NT3.51, and Mac OS since 7.6 and Mac OS X since the public beta. Maybe because my work covers so many domains and software I notice it more. If you just do a bit of word processing + email and internet then it doesn't matter which platform you use. I use combinations of Gimp, Hugin, Blender, Yafaray, VTK, Panda3D, OSG, Ardour, CLAM, Bristol, JACK, C/C++/Python, Gwyddion (for AFM data), Latex and latterly Mandelbulber almost every day. Each of these has replaced proprietary alternatives one by one - things such as Photoshop, AfterEffects, 3DS Max/Maya, MOTU Digital Performer etc. Usually I've tried open source software while waiting for funding or orders for proprietary software to turn up. It used to be the case that the interface was rarely as comfortable as the Windows/Mac OS X alternative. But don a hair shirt, and the benefits in stability outweigh a pretty interface. This isn't actually the case any more - Ardour, Blender 2.5 actually have better interfaces than their paid for counterparts.

The thing I've noticed is that open source software only gets better, whereas with proprietary software each version is a crap shoot - depending on how busy the marketing department for Adobe, MOTU, MS etc have been, each version will have more features, which may or may not be useful, but will generally introduce new bugs, make interfaces harder to navigate, and generally make software worse! Just to satisfy a revenue cycle.

...And that's before you start with support. Most community interaction I've had has been excellent, people have even added features just for me. You'd have to have a big turnover to get that out of Adobe or MOTU.

I'm not really an open source evangelist, but I'm becoming one. I came for the price, but stayed because it was better...

Blender 2.56 + OceanSim + Yafaray

2011-02-08T23:55:00.001Z

I just managed to get Blender 2.56 + OceanSim + Yafaray going. Things are really coming along:

Blender 2.56 Ocean Sim + Yafray from domejunky on Vimeo.

Dome Mandelbulber

2011-02-02T23:20:00.005Z

I've seen a lot of fractals in domes. Pretty much all of them have been 2D, with no dome correction, which I find quite nausea inducing, and definitely non-immersive. I was revisiting some sites I'd found looking into genuine 3D fractal rendering. Cool things were happening before with GPUs, but on this visit I came across Mandelbulber - looking through the manual I came across a reference to a 'fish eye' option, and got quite excited...

This is what Mandelbulber produces if you set the FOV to 1, and give it a 1:1 aspect to render to. I thought this could be an Equirectangular projection. Coupled with the fact that there were references to 'Spherical Mapping' in the source code. Remapping to fisheye gives pinched polar regions:

From this I assumed that it was a Cylindrical projection. I tried remapping from Cylindrical to fisheye:

This still doesn't seem quite right - I haven't tested it on a dome yet, but it just looks wrong...

[edit 18th Feb]

Well, 'Buddhi' the author of Mandelbulber has come up trumps - he modified the code to produce proper equirectangular output. So look out for fantastic 3D fractals in a Fulldome theatre near you...

IP over FC

2010-11-22T20:17:00.005Z

A few notes about my experiments with IP over FC.

We have Windows, Mac and Linux clients. We have qlogic fibre switches and mostly qlogic HBAs. On one of the vendors visits to fix the SAN, all the LSI HBAs were pulled from all the Macs, and replaced with qlogic. So during my research, I've been mostly hacking with these LSI cards. These are interesting from a Linux point of view since the drivers are all in the default kernel on just about every distro I've tried - including the IP over FC driver 'mptlan.ko' - these are part of the whole 'Fusion' driver.

When I first plugged these cards into some spare machines - up popped the network interfaces as fc0 etc, and standard 'ifconfig' got them all working. Before long I had NFS working quite nicely over the connection. But the interfaces come up as:


fc0       Link encap:16/4 Mbps Token Ring  HWaddr xx:xx:xx:xx:xx:xx
        inet addr:10.0.0.1  Bcast:10.0.0.255  Mask:255.255.255.0
        UP BROADCAST RUNNING  MTU:13312  Metric:1
        RX packets:72250 errors:0 dropped:0 overruns:0 frame:0
        TX packets:66105 errors:20 dropped:0 overruns:0 carrier:0
        collisions:0 txqueuelen:8192
        RX bytes:15940069 (15.9 MB)  TX bytes:564886898 (564.8 MB)

...and this 16/4 Mbps Token Ring, accurately describes the performance I'm getting. I've tried all sorts of things to try and force the card into a different type of encapsulation, but nothing works. I though it might be because I was connecting two machines point to point - but the same things happens if I bring up the card when it's plugged into a switch.

So I've tried to repeat the experiment with the qlogic cards we have. I've managed to get the driver from before the IP over FC was removed and at the moment I'm trying to find a distro that is close enough to RHEL or SLES to get this compiled, while at the same time being up to date enough to support the software stack I need to run on top (Realtime 3D/Video/Audio software) - but I have concerns about the practicalities of having to deploy this on visualisation clusters, and the brittle-ness of relying on out-of-date drivers...

So I'm wondering what my next move is - do I tell the University to jump ship to StorNext (another unknown quantity); treat the arrays as JBODS and let each machine do what it wants with its slice. Or pay for some contractors to come and re-configure MetaSAN to work (I have no faith in the product now, and a couple of days of hacking has taken the mystery
out of FC, to the point I'm not convinced there is anything left to configure)...

multi-projector Mplayer

2010-11-14T14:24:00.003Z

I was looking at cave-mplayer last night. Wondering if I could face patching the mplayer sources, and installing onto multiple machines, when I came across this link - which suggested that someone was trying to check-in a similar feature - synced multiple instances of mplayer over UDP.

I followed the thread to the end, to make sure it had actually been checked in.
Checked out the mplayer source from SVN - a bit of 'apt-get build-dep mplayer' brought in everything I need to compile apart from 'yasm'. And what do you know it
works!

wayland=homeless?

2010-10-29T14:31:00.002+01:00

I've been a bit concerned about Canonical's announcement that it was moving to Wayland. Nursing machines through the pulse-audio adoption was hard - particularly for multi-channel audio. I'm sure it'll be fine in the end, it's the interim that worries me.

We had an engineer from IBM visit us to install some Websphere technologies - he was a SuSE guy. I've never met someone who was serious about SuSE - I cut my teeth on SuSE back around '98 - before the Novell days. Ironically, I was looking after a few Novell Netware boxes at the time - so Novell aren't an evil firm for me - just serious engineers.

So I've been doing the rounds again - looking for a distro to base my development work on. Maybe more importantly, something that I can deploy with customers.

I suppose what I'm after is something as reliable as RedHat, but not quite so out of date in terms of graphics. I want more conservatism than Ubuntu, but not too the point that I've got to compile my whole development stack - which is beginning to teater:

For audio I need:

Jack + FFado
CLAM

For visual stuff I need:

Blender3D
Panda3D (with threading enabled)
Mplayer
OpenSceneGraph
VTK
VP8 and DNxHD codecs and support for VDPAU

I have to compile Panda3D to enable a more sophisticated threading model. With Ubuntu I was able to create my own .deb, which could then be installed on other machines in the cluster - but that's down the Panda3d source.

Ubuntu is quite a maligned distribution - people think of it as a pretty Debian, used mainly by people new to Linux. I know a lot of people however, who know their way around many distributions, who use Ubuntu as a way to simplify their administration of Linux, leaving them more time for actual development.

I've taken a look at Fedora over the years - but it seems even more bleeding edge and brittle than Ubuntu. Debian and RedHat(CentOS) for different reasons are to much like hard work to install and maintain the software I need. It also looks like RedHat/Fedora are heading for Wayland too. Just of badness a gave Gentoo(Sabayon) a go - I used to be a die-hard Gentoo user.

The surprise was OpenSuSE. It feels like a serious, rpm based Ubuntu - all my software was available in the repos. Customers who want support can look at SLES (although I'm sure Ubuntu would provide similar levels of support).

So, all these changes aren't meant to hit till the latter part of next year - but I'm ready...

From Ridley to Ragley

2010-09-17T01:22:00.003+01:00

I've been out of the saddle for over a month. Looking at a pile of new Capagnolo components on my desk.

It all started with buying a frame for my Girlfriends birthday. Getting rid of her MTB, and trying to get something specifically 'cross'. I ended up getting her a Paul Milnes frame - a lovely man from Bradford who sells cheap cross stuff via Ebay. The idea was to put the components from my Ridley onto this frame, and upgrade some of mine. The Paul Milnes is an amazing bargain. He'll sell you a frame and full carbon fork for £265, you can buy the same bike branded as Columbus for £330 or you can buy it as an Empella for £600 - it's the same bike, personally the pure white of the Paul Milnes looks way better than the others.

So that was the plan. I built up the Paul Milnes and she was off...I fitted some new Kinesis forks to my Ridley - with a view to raising my bars a touch. Raised the bars, and found I needed the saddle up a touch...oh, it won't move...and so began weeks of scratching my head, making tools, making spreadsheets etc. My carbon seatpost had bonded to my aluminium seat-tube. I tried brute force, WD40, cutting the seatpost from inside to out al la Sheldon Brown. I tried drilling, reaming. I eventually took up to the engineering block at Uni. The 'metal' guys didn't want to touch it without tungsten-carbide or diamond tools. The composite guy however was happy to put it in his big kiln to see of thermal expansion would do the trick...it didn't. He did find that heating to 130 degrees turn the resin in the carbon fibre to mush for about 10 minutes, allowing you to gouge out chunks of it.

This took a couple of weeks, and given the age of the frame I decided I had to stump up for a replacement. So I looked around and decided that a Paul Milnes was probably the best value around - It seemed a bit twee for both of us to have the same frame, but I'd get over that. When it arrived, I offered up the bottom bracket - a Campagnolo UtraTorque. Campagnolo are very proud of the small 'Q' factor of these cranks - the distance between the pedals. They don't like frames with heavily manipulated chainstays though - like the Milnes/Empella. Luckily Paul Milnes was a gent about taking the frame back. About a week later I went to a local cyclocross meet. One of the guys there was just starting a business importing A:XUS bikes from Germany, I sat on a couple - they are sweet. Tight angles, extremely light triple butted frame - they had one I could have for £499. The only other thing I had on the horizon was a 2009 Ridley Crosswind from a German website for £430. I was just about to press 'buy' on the Ridley, when I saw a shop soiled Ragley for £250 - They'd been on my list, but I thought they were expensive for what they were at £450. After a bit of research I decided to go for it.

So I've just built it up, and my initial impressions are that it's being undersold. It is triple butted, the angles are clever, and the whole frame has been cleverly hydo-formed. The UltraTorque crank fits too...

Dome VP8

2010-07-26T13:53:00.008+01:00

Doing some prep for the panotools conference coming up next week. We have some requests for the ability to view Equirectangular video - as supplied by cameras such as the Ladybug. I don't have access to the SAN for movie playback so I've been searching for a codec/container combination that will let me playback 2048x1024 video on a 3D texture inside Panda3D. H.264 is a bit CPU intensive - so I thought I'd try VP8...

I downloaded that latest nightly build of Chromium - this gave me a working libvpx. I compiled ffmpeg 0.6 against this new libvpx. I ended up with using these settings:

./configure --enable-shared --enable-nonfree --enable-pthreads --enable-libfaac --enable-libmp3lame --enable-libvpx --enable-libx264 --enable-gpl --enable-pic --enable-libdirac

I then compiled Panda3D against this copy of ffmpeg. This didn't work initially. Well, it might have worked - but VP8 is limited to certain video containers, with Matroska being the preferred. So a quick 'egrep' in the Panda3D source directory yielded:

./panda/src/grutil/config_grutil.cxx: ts->register_texture_type(MovieTexture::make_texture, "avi mov mpg mpeg mp4 wmv asf flv nut ogm");
./panda/src/grutil/config_grutil.cxx: ts->register_texture_type(FFMpegTexture::make_texture, "avi mov mpg mpeg mp4 wmv asf flv nut ogm");

So I added 'mkv' to the list of know container types - and it worked. So now I have VP8 inside a Matroska container, applied to a 3D texture. So far the performance seems remarkably good.

Panda3D with VP8 Matroska from domejunky on Vimeo.

Atomic Scale...

2010-07-13T03:49:00.005+01:00

Just recovered from fulldome.org.uk - but no rest for the wicked, we have an Atomic Force Microscopy workshop coming up

My job is to find some data paths from sample -> dome, and maybe even sample -> sonification, to scaffold the workshop for some Arty/Hacky types.

This is the first pass at some data from our AFM, simply using the bmp as a height map, into blender and into the dome via the Game Engine.

AIS Tracking

2010-06-27T01:09:00.011+01:00

Pixi-->GPX

2010-06-14T01:16:00.005+01:00

I've been having a totally Python-tastic weekend. Rode up to Dartmoor over the weekend. First time using 'Mapping Tool' - on a Palm Pixi.

Mapping Tool has a super-slick feature, it lets you select CycleMap from openstreetmap.org as the renderer, and it will show you all the national cycle routes. It also has another feature, which is less than cool. If you want to export a gps track, you do it via email, or multiple emails if you track is deemed to large. Our trip to Dartmoor warranted 10 emails - the Pixi gives up after 4 or 5.
I decided to get the gpx files off the phone myself, like I used to do with TMJ (Track My Journey). What I found were hidden directories with SQLite databases, three in total. I have about 6 tracks on the Pixi at the moment, so I couldn't work out why only 3.

I've never touched databases in Python. C, Objective-C, Java, Perl, Ruby - yes, but not Python. The road seemed to rise to meet me again, and I quickly had the data out of the database - and worked out that all my tracks were in one of the databases - with differing IDs.

Next was creating GPX style XML. Parsing XML with Python I'm familiar with, but not creating - last time I had to use 3rd party libraries - 'XMLTramp' I think. A bit of xml.dom.minidom and I had compliant GPX files that Google Earth would accept...

DomeStreetView

2010-06-12T00:33:00.003+01:00

I was asked on Friday if I could get the tiles out of Google StreetView for a particular location

- they wanted to create some graphics for the event from them. A preliminary google found this - after trying the code in Processing, and being reminded what a poor environment it is - wouldn't even let me resize the editor window, might as well have been using it on a mobile phone - I ported the main method to Python, and managed to get the tiles to satisfy the original request. At Zoom level 3 Google give us a 4x7 matrix of 512x512 tiles. Not 3x6 as suggested here. From looking at the tiles, it looked like Google were serving up a standard 360° 2:1 Equirectangular projection. I manually composited the tiles into one 3584x2048 (7x512, 4x512) - and immediately, I could see that it wasn't.

What we have is more than 360°, and an image that has a black bar at the bottom, meaning that it isn't 2:1. So, by a process of trial and error I worked out the final image size of 3328x1664.

Now I had a compliant Equirectangular, I couldn't leave it at that, I had to see if I could get it into the dome. Took about 20 minutes to grab the bits I needed from my EquiFlickrAuto app, and it was done - StreetView in the dome! Now I need to get a decent way of selecting the location, maybe just start it from a particular location, and use a Wii controller to move through the Panos. The google maps API seems completely Browser/Javascript focused - although this guy has managed to find the URL based API....

Classic one-liner...

2010-06-08T01:30:00.003+01:00

for i in *.MTS; do ffmpeg -i "$i" -s 1280x720 -r pal -b 115000k -threads 4 -vcodec dnxhd -acodec libmp3lame `basename $i`.mov; done

Class, I'm using it to transcode AVCHD .MTS files from my GH1 to DNxHD in a .mov container. DNxHD is from Avid, very nice intermediatem cross platform CODEC.

Guerilla Armchair Cycling

2010-06-04T01:34:00.001+01:00

I've been spending most of may watching the Giro D'italia, and a bit of the Tour of California - I've discovered Twitter as a way to get the latest 'griff' on what the cyclists and pundits are up to.

And I've managed to transpose all this to my new phone. I Like WebOS - I didn't like the Palm Pre, and for a while things looked shaky at Palm. But with the arrival of the Pixi, and HP buying Palm, it all looked quite rosy - so I took the plunge. Back on a contract after two years of keeping O2 on PAYG Probation...

Dome Volumetric Fruit Fly

2010-05-02T01:51:00.001+01:00

We've received some DICOM data from a University in Austria - in connection with a project for the Biology department. A Drosophila fruit fly. I've just managed to get a path from the raw DICOM files into OSG. I had to compile OSG myself to get support for 'dcmtk'. I also had to use the '--mip' flag before I saw anything. I then modified our osgdomeviewer to accept the DICOM data and render a 180°x135° truncated fisheye. It worked...but I was only getting 2-6 fps. On a hunch, I checked out a new nVidia driver for my card. Up until now, it had been recognised as a GT230, rather than GT240 - and it had a tantalising 1700MHz mode greyed out. The driver upgrade brought the fps up into the 20s - not bad for a £59 GPU...

I'd like to put a nice mask over the image - same as I do with my Panda3D based dome applications, but C++ is so verbose...

...4096 - the number of the beast...

2010-04-08T01:36:00.004+01:00

I went through a bit of a 'mare trying to get the credit sequence out of a backup tar I had. Forced me to buy a new case (cooling) and PSU and a 1TB drive - which I'm going to start using as removable eSATA cartridges - at £57 for 1TB - far cheaper the RDX, and no restoration process - I'm not going to even bother making tarballs anymore...

...but the drive was one of those Western Digital 'EARS' Advanced Format drives. It clearly states on the packet:

"All other OS configurations - drive is ready for use as is"

Well, It looks like the firmware lies about the type of drive it is...and, older version of fdisk and parted don't know about 4k drives anyway. So had to download an ISO of the GParted live CD and boot from that - then repartition the drive with a GPT partition map, and a sector start point of 40....

I seems to be working. I was getting 60-70MB/s on a write operation!

...first footage...

2010-04-04T01:06:00.003+01:00

A week with the GH1 + Coastal.

Over Easter I went to 'planet fanet' to visit the aged p's. Took the GH1 + the Coastal 4.88mm and the Zenitar 16mm. Pretty dramatic clouds, good exercise in exposing for varying contrast as the sun popped in and out. I need a way to get Neutral Density filters behind the fisheye. The Coastal Optics has a fixed aperture of f5.6 - given that the sun will pretty much always be in shot in a fisheye, I was having to shoot with a shutter speed of up to 1/1000 sec. Even with the Zenitar on f22, I was on 1/250 sec. Luckily there is room in the shift adapter for an ND filter, maybe two. I tried a 52mm ND, and a 40.5mm Skylight - they're a bit big - I reckon a 37mm.

The process of editing the footage properly has thrown up some issues. I'd planned on using Blender, like I'd been doing with H.264 from the Sanyo - but Blender doesn't seem to like the Advanced profile H.264 from the GH1. I had some success with OpenShot. It liked the footage from the camera, and allowed simple cuts and fades, and the application of a mask to clean up the fisheye. When it matures it'll be something like iMovie in complexity - which is no bad thing. I started having issues with renders honouring the timing of edits and fades - probably down to the long GOP of the video. I had another look and Kdenlive - I have a love/hate relationship with Kdenlive. It's been less than mature up until now, and it relies on QT which makes the interface feel clunky. I tried my 1080p H.264 from the GH1 (PAL) and it loved it - playback was a bit sluggish, but OK. I also discovered the amazing DNxHD - it's offered as a transcode option in Kdenlive. I started transcoding all my clips to it - and the playback in Kdenlive becomes much better.

As I added a bit of shift and fades etc, Kdenlive started getting a bit crashy - seems the version shipped with Ubuntu 9.10 is a bit old. I upgraded to 7.7.7, and it's looking good. The following is and edit of some of the clips I got...

Coastal (optic) from domejunky on Vimeo.

...I made a purchase...

2010-03-29T00:56:00.005+01:00

so, I went and bought it. Took a week for the Fotodiox shift adapter to arrive. Looks like I have one with the low light vertical streaks - maybe they'll burn off. I spent 4 hours reading a 56 page thread on DVXUser - weird shit! I only spent £475 on the camera, a refurb body-only from srsmicrosystems.co.uk - who let me change from a GF1 when I found out it didn't have manual ISO (thanks to Philip Bloom!)

Here's the first (uncalibrated focus) shot through the GH1 + Fotodiox + Coastal Optics 4.88mm:

GH1 + Coastal Optics 4.88mm from domejunky on Vimeo.

...sweet spot...

2010-03-22T02:02:00.005Z

I spend a lot of my time researching camera and lens combinations. Since everything for the dome has to be at least part of a fisheye - that's where it starts for us. While other photographers and cinematographers might consider the fisheye an effect to be used sparingly, for the dome it's a pre-requisite.

It usually starts with a new camera, new lens, but eventually I end up looking at sensor sizes, 16:9 crop dimensions and the size of image produced by different fisheye lenses. So by way of an example I'm going to try and record it once and for all...

In terms of personal cameras, I use a Nikon P5100 + FC-E8 fisheye - really pristine fisheyes, and time-lapses. I also use an Xacti HD100 - the one that can only do 1080i - which I avoid, and use for reasonable 720p documenta of things in the University and my life, that I can edit in Blender....I've tried FC-E8 and FC-E9 fisheye adapters on it, but they crop extremely. So I've had my eye out for a compact camera that can do video with a fisheye adapter. Most compacts don't have wide enough lenses, and usually don't have a bezel - the P5100 was a bit of a find. The Xacti taught me that you can get decent video out of most things if you have full manual control - but you need a decent sized sensor. This is what lets our Canon HV20s, and the D90 down - they give you control over most things, but reserve the right to mess with the ISO or sensor gain. A good camera will let you take completely black video.

We got a Nikon D90 at work - I thought that good 720p was preferable to poor 1080p - well the D90 does neither, you can nurse the occasional good sequence out of it, through good glass - but it's not up to the ravages of a fisheye. Our Coastal 4.88 crops on it in video mode.

The Xacti taught me that you can get decent video out of most things if you have full manual control. This is what lets our Canon HV20s, and the D90 down - they give you control over most things, but reserve the right to mess with the ISO or sensor gain. A good camera will let you take completely black video. The P5100 + FC-E8 taught me that there's only so much accutance available via adapters, and that they rely a lot on the optics of the parent lens.

It was the Canon 550D this time - just been released, does 1080p, practically the same features as the 7D, but pretty cheap. I know from using our Coastal Optics 4.88mm, that this camera will crop the fisheye circle slightly. So it didn't seem like an option. While fishing around in indie film making forums, I found a lot of people talking about the Panasonic GH1, and how the 7D/550D didn't spell the death of it. I've never really considered Panasonic as an SLR manufacturer - but there are a lot of people using GH1s for serious cinematography - and the four-thirds format is becoming a hobbyist's playground.

I stumbled upon this shift(not tilt) adapter:

Pro shift adapter

...and suddenly whole new set of options opened up. A day of research and furious scribblings later I'd worked out the following.

Our Coastal Optics 4.88mm has an image circle diameter of 14.9mm. The GH1 has a sensor size of 17.3 x 13mm - a 4:3 aspect ratio. In video mode it uses a 16:9 aspect ratio. 4:3 could be expressed as 16:12, so the 16:9 mode uses 3/4 (three quarters) of the vertical resolution of the sensor - this turns out to be 9.75mm. Three quarters of our original lens image circle is 11.175mm - so it's not an exact fit, but close. The use of the shift-adapter means that the image circle can move until all the cropping is on one side, this should be a distance of 1.862mm (heh!) The Sunex 'super fisheye' which is 5.2mm with a 14.5mm image circle - or the 4.5mm Sigma, might work better, since they're a bit smaller....

...perhaps an image:

...More than meets the eye...

2010-03-06T00:47:00.024Z

Prompted by a visit from Global, where some of my real-time applications on our Skunkwerks box failed to launch, I've been doing a bit of bit-rot fighting.

I got all the OpenSceneGraph stuff working. Furnished with confidence and knowledge from a fortnight and a gig hacking Fluxus and Scheme, I decided to add a 'camera tickle' feature to our Panda3D Flickr Equirectangular viewer. This 'camera tickle' was just a bit of sin/cos based movement on the fore/aft left/right of the camera. Inspired by a conversation I had once with D'nardo from Elumenati. The idea being that the movement would stop the image 'flattening' onto the dome surface.
...well that turned out to be a can of pythons

While trying to ascertain the effect/affect of the 'camera tickle' - judging it by the amount of vection induced - we realised that the panoramic images weren't 'unfolding' properly onto the dome. I tried hack the variables in the code, some were better, some worse, but none were right. In particular, objects that should appear vertical as they move behind you, were laying down. One thing we did realise is that tiny adjustments in dome to fisheye calibration have a huge effect on vection induction.

So the code was a bit crufty, and I'd been hacking on it since sometime in 2007/8 - so I decided to rewrite it properly. I made my fisheye script into a proper class. Got rid of the invisible RoamingRalph. I changed the code so that it didn't create a sphere per equirectangular file it loaded, rather it changes the texture on a single sphere. I also remodelled the sphere, to use more polygons, and reused it as the cubemap sphere.

Things were looking good, we loaded it up in the dome...and still had the same issues. Now, with the 'knobs' for camera position and tilt cleanly exposed, the issues were more obvious. So a good couple of hours of head scratching, hacking, reading and modelling. I decided to go back to the old testament.

Paul Bourke

Now, my approach had been to use a dynamic cubemap, mapped to a sphere, with an orthographic camera looking at this. I culled the front face of the spehere - a hangover from testing the approach with perspective lenses, where it makes a difference. My approach produced a fisheye like this:

...You should be able to notice that the concentric circles get closer toward the edge. Compare Paul's render of a Equiangular fisheye:

...Here the concentric circles are the same distance from centre to edge. Suddenly it made perfect sense - a threshold concept moment. Of course an orthographic camera would get narrower towards the edge. The effects can be mitigated a bit by using a perspective camera. But as you can see from Paul's site, the variation in concentric ring spacing is shifted to somewhere else between the centre and edge:

This was taken with a 60 degree camera.

This was a 90 degree camera.

So, close, but no cigar...

This was quite a blow. I have a lot of code written using Panda3D, and these subtle fisheye differences make big differences to induction of vection - so as a platform to base a series of presence experiments on, it was looking shaky. Panda3D has a fisheye class built in to it - I think Elumenati were involved with Carnegie Mellon's ETC, and helped them develop it. But I'd never got it going...

http://www.panda3d.org/phpbb2/viewtopic.php?t=1264&highlight=fisheye

I think I was suffering from poor Panda3D/Linux interaction at the time. I tried again with the code, and got it to work. About 8 hours later, I'd managed to get the geometry correct for our 4:3 rear-truncation, applied a mask, and had my old code running on the new class. I had some issues with the interaction of setFimSize() setFilmOffset() and our 25 degree tilt - instantly you can see that the fisheyes are equiangular though.

This is what the finished item looks like:

NLEquiAutoFlickr from domejunky on Vimeo.

Aggregator

2010-02-27T14:39:00.009Z

I've just finished a 4 hour livecoding performance in the dome. Using fluxus and the following live sources.

PALAOA Audio Obsservatory (microphone under ice)
http://icecast.awi.de:8000/PALAOA.MP3

Air Traffic Control:

http://mso.liveatc.net:80/khnd1
http://aus.liveatc.net:80/sbbr_acc

Calm noises:

http://www.whitenoise247.com/Sounds/CalmSeaWaves.wav
http://www.whitenoise247.com/Sounds/river_full.wav

Natural Radio:

http://mp3.nasa-us.speedera.net:8000/mp3.nasa-us/florida1

http://67.207.143.181:80/vlf1
http://67.207.143.181:80/vlf3
http://67.207.143.181:80/vlf9
http://67.207.143.181:80/vlf15

http://194.116.73.37:8000/pontese124.m3u
http://icecast.nis.nasa.gov:8000/florida1
http://picasso.astro.ufl.edu:8000/icy_1

Radio Astronomy

http://28.72.128.252:8000/radast

I like fluxus, I like the immediate mode style programming. I don't know if I can get on with scheme though. This is what it looks like:

;(require fluxus-016/drflux)
(require fluxus-017/planetarium)
;(set-dome-mode! #t)
(smoothing-bias 2)

(clear)
;(clear-colour 0)
;(blur 0.1)
;(fog (vector 0.1 0.1 0.1) 0.2 0.01 0.1)
(ortho)

(define dome (dome-build 10 180 2048))

; buffersize and samplerate need to match jack's
(start-audio "MPlayer" 1024 48000)

(define (render count)
(cond
((not (zero? count))
(translate (vector 0.1 0.1 (* 10 (gh 4))))
(scale (vector 2 2 1))
(rotate (vector (gh 4) (gh 5) (gh 6) ))
(colour (vector (* 0.5 (gh 4)) 0.2 (* 0.5 (gh 10))
0.3))
(opacity 0.3)
(draw-torus)
(render (- count 1)))))

;(with-state
;(rotate (vector 0 -25 0))
;(render (- count 1))
;(draw-cube)

;set the view of the camera
(dome-setup-main-camera 1400 1050)

(every-frame
(with-pixels-renderer (dome-pixels)
(with-state
;(rotate (vector 0 0 (* 90 (cos(/ (time) 10)))))
(translate (vector 0 0 -100)) ; move it into view

(render 10))))

Asynchronous Hell...

2010-02-05T21:03:00.006Z

I've just been on a week long hunt for a method of getting data from live feeds into Blender Game Engine. Not to taxing a task you might think...

I put together a simple example for a workshop we've just run. I wrote a python script in Blender - I grabbed the XML data feed from the Arch-OS system. used 'lxml' to grab the value of bms_.WindVane - the wind vane on the top of Portland Square. The idea was to have the camera in Blender align itself with with the wind vane in 'real-time'

Problem is, that querying the XML script takes from 0.5 to 4 seconds - and the 3D refresh cycle was blocking for that amount of time. I had a fiddle with threads, which didn't really yield solution. I asked around on #blendercoders, and was pointed in the direction of Python process, multiprocess and subprocess modules - the latter of which I'm quite familiar with - I've been using it to kick-off movie playback on remote machines using our AMX Panel.

I now understand process, multiprocess and subprocess, and threading a lot more now - but still didn't get any closer. The issue is the communication between the threads or processes, even if using 'queues', is the part that blocks the 3D.

I thought It might be a case of needing to run the slower processes in another Python instance...made sense - but it seems even the sockets used for communication either block, or if set to 'non-blocking' go out of scope...I think a similar happens if 'join()' is ommited.

I did finally get it to work. OSC.py came to my rescue. I figured that CLAM use OSC (and SpatDIF) - to loosely couple their audio backend to BGE - I've done it with Panda3D + CLAM, and BGE + CLAM with 'Yo Frankie'. It did work...but only in combination with a subtle python script I found for Blender:

if not own['connected']:

#print "Hallo blender Welt"

own['connected'] = True

GameLogic.socket = socket.socket( socket.AF_INET, socket. SOCK_DGRAM )

GameLogic.socket.bind(('', 1234))

GameLogic.socket.setblocking(0)

GameLogic.socket.settimeout(0.02)

The difference here is that they've created a persistent, non-blocking socket - which I assume is acting as some kind of buffer for the OSC messages...

Kudos to this guy

EuroGraphics 2010

2009-12-09T15:45:00.004Z

I'm writing a paper for Eurographics 2010

- something like: 'Towards an academic praxis for domed virtual environments' - Slub and Wrongheaded get a mention, forming part of this praxis.

Anyway - they need full ACM formatting, and the templates are all in Latex. They provided instruction for converting from various types of text.

I don't really get on with OpenOffice, just feels a bit bloated to me, so I use AbiWord. I got to a point in the paper, where It wouldn't let me past in a block of text from another paper - this other block happened to have a quote in it, which itself had a quote. Everytime I pasted the block, Abiword would crash. OK, tried another machine, same. OK, found a bug report about it, compiled myself another version, same. OK, let's use OpenOffice, same! Things were getting really weird, I managed to get the document into Google Docs, and that was OK - but Google docs doesn't feel sustainable for my PhD.

So today I did what I do abut every 18 months, and went to have a look at the Latex world. I was wondering if there were any new GUIs that would get the job done. I found a thread where people were discussing Lyx, Kile, Emacs etc - the usual suspects. Then right at the end of the thread a guy suggested the latex plugin for Gedit. I installed it, and I'm a happy guy now.

I'd been thinking that programmers have much better text processing tools than authors - I was even thinking about writing my PhD in python - just to get the mark-up colouring, then just have a uber routine that calls all the chapter, and paragraph methods...

But this seems to be similar to that but better. I spent a couple of hours making sure I could get my docs into this ACM format. Learnt half a dozen Latex commands, so I can quote and section text, and I'm away. Bibliographies are a doddle! After wrangling with all these templates classes, font definitions, styles supplied by ACM. Their approach was for you to use the \include command to to bring your text into their document structure. I've worked out that you can do the opposite. Write your Latex document, then have the first line include all their stuff, really clean.

Yafaray + Movies

2009-11-23T20:08:00.003Z

Yafaray just got movie textures back! Very useful for me, let's me do all the things I'd have to use AfterEffects for, and gives me more control.

Not in the main builds yet - but it is in the darktide branch:

svn://svn.yafaray.org/repo/yafaray/branches/darktide

...remember to do 'scons swig' after the main scons build t build the bindings.

...D90 + Coastal Optics

2009-10-15T04:33:00.005+01:00

Nikon D90 + Coastal Optics 4.88mm from domejunky on Vimeo.

Our D90 arrived to today

- I've modied the t-mount to raise the lens slightly as in movie mode the fisheye is truncated top and bottom. This way we nearly get our 4:3 truncated fisheye. I'm a bit disappointed really - I thought the exposure would be truly fixed in manual mode when recording video. It isn't, you can fix it by holding the 'AE-L' button, but without you see it stepping between ISO or shutter speed settings - maybe both. Not like the little Sanyo, if you want to record black video it'll let you.

But it seems OK if you chuck loads of light at it. It doesn't talk to Nikon Capture Pro - so dialing in the focus on the fisheye is a bit more laborious. They are solid little cameras though, not as cheap feeling as I remember. I think it's a good time to let the waters settle in terms of SLRs with video - The Panasonic GH1 is getting a bit of a following:

http://www.hotrodcameras.com/

...as is the Olympus E-P1:

http://funahara.com/blog/?title=a_19981_a_23450_a_26399_a_26356_a_26032__65298&more=1&c=1&tb=1&pb=1

...I think the four thirds standard might stick. It's interesting to compare sensor sizes to film:

http://www.hotrodcameras.com/2009/06/17/film-format-and-sensor-size-comparison

memoryade...

2009-10-15T04:07:00.001+01:00

mplayer dvd:// -vf crop=480:360:120:120 -vo vdpau:denoise=0.6:sharpen=0.3

...just so I don't forget - plays a DVD and crops for truncated 4:3 projector, and uses VDPAU to clean it up a bit.

interlace...

2009-10-11T00:58:00.003+01:00

I've just got an old Miro DC30plus card going under Linux

- I need a better TV, but don't want to buy a dedicated one.
I had to put 'options zr36067 card=3' into '/etc/modprobe/local.conf' to get it recognised properly. Once it was working I found I was getting jerky video. A bit of sniffing around with mplayer and I discovered that I have interlaced, bottom field first video coming in off the card. I started playing with a few de-interlacers - then discovered yadif - I realised that in one of modes you can set it to produce a complete frame from each of the interlaced fields - coupled with a framerate of 50fps (double 25fps for PAL) I can playback interlaced video on a progressive video display, without actually using a deinterlacer, in effect you perception is doing the deinterlacing - same as an interlaced TV.

This got me thinking about an interview I read with James Cameron. He was urging Hollywood to move to 48fps (double the current 24fps-ish) - he was convinced that it doubled the perceived resolution, without quadrupling the size of you production and delivery pipeline...

We're running 30fps movies in our dome on 60Hz projectors - like most domes do, I've decided to try some experiments with my new friend 'Yadif', see what happens it I run 60fps interlaced on them....

Fulldome Fluxus

2009-10-07T16:22:00.002+01:00

Prompted by seeing J Walt's 'Omnicentric Universe' at Domefest09 - only a few days after hosting a 'Livecoding' event at Plymouth's dome - I've been having having a play with Fluxus. It's a very compelling environment, actually quite fun. I'm not sure about scheme yet - too many parenthesis for me. I imagine you get used to it - I used to hate the white space in Python too...

Fulldome Fluxus from domejunky on Vimeo.

Livecoding at the Dome

2009-09-21T02:15:00.002+01:00

Just got back from hosting Matt Yeeking et al for what maybe the first Livecoding event in a Dome...? Hardwork getting 8 channels of audio mapped back out into 10.1 - think we blew a speaker, again...plus VNC session for each performers machine being fed to the QXGA Zorro - lot of network traffic....

Audio recording didn't start - pressed 'on' in Bidule, but didn't press 'start' - need to ween myself off the RME totalmix interface for routing - I could have used JACK, and then Ardour to do the recording....have to make do with the audio from the video...

KVM on the Blade...

2009-09-05T12:23:00.001+01:00

I'm revisiting our attempt to get an AferEffects/3D Studio Max render farm running.

The problem is that 32bit AE can't see more than 2GB of RAM. We have 8GB and 8 Cores per blade. My plan was to put four virtualised instances of XP onto each blade - giving each instance of AE 2 cores and 2GB RAM.
I tried the Citrix Hypervisor, but they want $$ for the client application - so I thought I'd give KVM another try. Before, when I went through the various Hypervisors, I was getting correupted video and sessions timing out. Reading around suggested that a firmware upgrade to the chassis (an M1000e) and to each blade would fix this. This has to be the scariest firmware update I've ever done...The chassis has 9 high speed fans on the back which, when up to speed, sound like an F16 - these came on full blast during the upgrade, and for about 5 minutes I thought I'd bricked the whole thing....

...So after the update, no more timeouts or garbled VNC. XP went on fine. When it came to installing AfterEffects though....hmmm

We have AE on a 5 disk installer (Adobe production Premium), disk one was recognised and AE started the install. After a while the installer asked for disk 2 - I tried everything I could to eject the first, and insert the second. I could get the disk ejected, either from XP or from Virtual Machine Manager. But nothing I tried could get the second disk recognised. I even went and had a chat with the KVM developers via IRC - they suggested using the direct libVirt console - but that had the same behaviour....

I think I need to find a 64bit version of AE, and forget about Hypervisors for a while...

Dome audio fixing

2009-08-15T02:14:00.002+01:00

I finally fixed the audio in the dome this evening. Tracking the issues down was hard enough. Three different 5.1/7.1 sources, a handful of stereo sources. Switched through an Extron 12:8 matrix (with an evil manual). Feeding two RME Firefaces, which via a physical modelling simulation, feed the 10 actual speakers.

Once I'd found a way to get a signal to a know channel in the Extron, is was pretty easy really - they're amazing devices, can even program in A/V delays. The rear has more BNC sockets than you've ever seen, so densly packed that they give you a special tool (that looks like a tool for taking test cores from cheese) to fit them...

Blog Transfer

2009-08-13T01:42:00.000+01:00

Sweet! Just managed to get my blog moved from my old self hosted Wordpress site, to this blogspot site.

I tried a few tools, a Java based tool called 'blogsync' from a guy call Yichao Zhang - it worked, but didn't import the body of the posts, just titles. Probably because of the old version of Wordpress I was using - which is why I didn't have an export button in the first place.

The second thing I tried was this:

http://www.cuberick.com/2007/05/migrating-from-wordpress-to-blogger.html

It took a bit of hacking, but it worked finally. So I have all my blog entries going back to 2005.

Lenovo + MultiFace

2009-07-28T01:31:00.000+01:00

Well, the focusrite didn't stay. Too much unreliability. I don't know if this is FFADO, LibFW or JACK. I've never had a super-reliable firewire based device under any OS. Nothing like a PCI Delta 1010 under Linux...

...Until now, I shelled out for the ExpressCard version of the RME Multiface. Runs 10 channels (S/PDIF, and ADAT) into the RMEs in the dome, no problem. CLAM and JACK love it. The Realtime kernels from Ubuntustudio suddenly became stable too - even with evil 3D going on.

One side effect is the insanely low latencies you can get with RME - I started playing Guitar more. Through and new discovery - Rackarack, like GuitarRigg for nerds, but cool.

I also just invested in a Netbook. A Lenovo S10. Bought to keep all my email and PhD etc separate from my worstations...it's working quite well so far. I'm trying to get it going with the Multiface (The Lenovo has an ExpressCard slot)

We even used it for a bit of SatNav on a recent trip to france, that would have been more succesful if we'd found some decent WiFi to download the maps.

ffado, firewire and focusrite

2009-06-05T22:00:00.008+01:00

What a nightmare. Looking for a new audio interface to drive the 10.1 system at work.

To do this I need to use ADAT + 1 S/PDIF - all the analogue inputs are taken on the RMEs at work.

I'd been playing with the MOTU and a version of FFADO built from the latest trunk in subversion - with some success. I say some - the digital clock in the MOTU seems very flaky and easy to upset. This is the second one I've had - and both were the same, even under Mac OS X and XP. I've tried quite a few firewire cards too. Between the machines I have here I had a Via, 2 Agere, 2 Ricoh - so I went out and bought an ExpressCard one that had a Ti chipset. For the record (it took me a fair bit of research) the Cardbus and ExpressCard firewire devices that maplin sell are based on Via AND Ti chipsets - you need to look at the model numbers to find out which - mine has a number of CE-T08-2210......... - the important thing is the 'T' the Via ones have a 'V'. The new card did make some difference, but as soon as I started and 3D process I'd get xruns and gaps in the audio, or jack would crash.

So I thought I'd work backwards from the ALSA, FFADO lists of compatible devices - I need 4 mic preamps with phantom power for the TetraMic; 4 balanced outs for a quad setup at home (in my bedroom); and S/PDIF and ADAT for driving 10 speakers at work. I don't know if that was a lot to ask, it's not so easy to find. I decided I'd settle for 2 out of three and forget about the preamps - I could also nurse the MOTU for that.

My mate Matt (who is a gigging, signed musician using PureData, Ardour and Supercollider under Linux) uses an RME Multiface + ExpressCard. I rate RME, and was aware of their solid Linux support. So that was the main contender really - the only problem is that they're about £750 in the UK at the moment - his cost him around £300 some time ago, so that's easy for him to say.

Then I spotted the t.c. electronics Konnekt 24D on a database of devices as having ADAT. It is based on t.c. electronics DICE chipset. I did a bit more digging and came up with a few more devices based on the same chipset, notably the Focusrite Saffire Pro40. I noticed on the FFADO blog, that support for the DICE based devices was coming along, and that Focusrite were being Linux freindly. Even donating a few interfaces to the FFADO project - I sat around in the FFADO IRC channel for a bit, and discovered that DICE support was actually already there - just the GUIs to the devices features weren't working. I was left with a dilema - swallow the £750, and be able to crack on the next day or buy a DICE based device and maybe be in for months of hacking and instability. In the process I found a B-Stock Konnekt 24D for £190 - "I can live with that" I thought - and went off to order one - I thought at that price I'd have to take a plunge.

When it turned up, it transpired to be a really nice device. A massive improvement in sound quality compared to the MOTU. It turned out to be pretty robust too when running with FFADO - suddenly I went hours without restarting JACK. I was doing meaningful work in CLAM! I quickly modified all my CLAM projects from quad to 10.0 and ran up to my dome to see if it would work...

...It didn't - I got audio into the RMEs, but accompanied with an awful 'I can't sync' Geigger counter noise - through every channel...ouch! I methodically tried all sorts of syncing options. When I got the interface home, I tried it with the MOTU - worked like magic. Back in the dome I tried the MOTU with the RMEs just to check I hadn't been dreaming when I set it up before, worked like a charm. What was more, I could daisy chain the ADAT from the Konnekt 24D to the MOTU to the RMEs fine. I searched the RME and t.c. electronics support forums, sent emails to the support departments with practically no feed back. I sent the interface back and considered my options.

Returning the previous device had been pretty easy with Digital Village's 7 day no quibble - so I figured that maybe a should give the Saffire Pro 40 a go, and just return it, and stump up for a Multiface if it didn't work. I wasn't totally convinced, so in a last rash attempt to get this solved, I phoned up Focusrite - they're in the UK, I should at least try to take advantage of that. I got put through to a testing engineer, who spoke to me quite honestly for 30 minutes about the DICE chipsets, ADAT implementation. Most importantly he said they would have repeated the Saffire Pro 40 --> ADAT --> RME combination in their labs and with their Beta testers. Seemed pretty fair to me so I placed the order...

...It arrived today! It worked with FFADO after a firmware update - I had to boot it up attached to a Mac to do this, and to set the routing in their mixing GUI. Back into Linux and solid as a rock - same liquid sound as the Konnekt -so maybe it is the Jitter suppression that everybody goes on about. I took it to the dome and plugged it into the RMEs, and out popped 10 channels of lovely clean audio - sweet! I couldn't believe it was so easy - I've had it in stress test mode all evening - listening to Ambisonic recordings, and pumping the audio through the ADAT out connected directly back into the ADAT in, 4 times! It's a good was to test, as you don't need 10 monitors to hear any artifacts. It has a killer feature for multichannel work - any combination of the analogue outputs can be assigigned to the volume knob on the front.

So I'm a happy guy now, and a new Focusrite fan...but it made me think, when did I last buy something English (even though it's made in China).

EquiAutoCLAM

2009-05-11T13:34:00.003+01:00

I've been playing with CLAM - thanks to some help from their Dev team, I got Blender talking to CLAM on a different machine using the SpatDIF protocol. To test the technology I decided to create a rotator for equirectangular images that would trigger a CLAM network to rotate a matching Ambisonic recording. After a bit of head scratching over the difference in scripting between Blender and BlenderGameEngine - I got some signalling between the two.
In A flash of inspiration I decided to try and get Panda3D talking to CLAM - since the equirectangular code already exists. I moved OSC.py to the working directory, pinched a function from CLAM's test python, and off I went - amazingly easy.

This is the result:
Machine A: Running Panda3D

Machine B: Running the CLAM network

Dome Corrected BGE

2009-05-09T18:11:00.002+01:00

I've been playing with the latest build of Dalai Felinto's dome corrected Blender Game Engine. This is one of the levels from Yo Frankie. Performance still a bit suspect, even running on Quad Core laptop with Quadro 3700.

We get some stitching errors on reflective objects. If the FOV is increased to 181 degrees, it uses 5 rather 4 tiles, so the stitching is less noticable, if not invisible. Next step is to try and integrate this with CLAM

BGE in the Dome from domejunky on Vimeo.

EquiAuto

2009-04-26T23:26:00.000+01:00

I've had a request for a more approachable way to display sequences of Equirectangular imagery in the dome. So I've revisted the code I wrote in Panda3D a while ago. I tidied up my Fisheye class - with the help of some of the Panda3D developers I managed to track down a really weird bug I'd been seeing on and off for ages. I swapped the Perspective camera for an Orthographic camera.

This particular app looks for all the png, tga, jpg equi's in the current folder and rotates and fades its way through them....

EquiAuto from domejunky on Vimeo.

2009-02-22T22:08:00.000Z

I did it! The bottom bracket tool arrived saturday, and I completed the building of the bike. I had a kind of dry run first - I was worried about the threads in the BB - I just needed to know I wasn't going to be thwarted at the last minute. I went out and bought some threadlock, grease and silicon gloves, and took it al apart and did it properly. Nothing about the assembly was particularly difficult - I think the tolerences tell in the longevity - I hope I fair better than the average on the internet, roughly 300 miles.

Just for posterity: The crank was an FSA Gossamer. BB Cups were torqued to 30 ft-lbs. Axle-cap was torqued to 6 in-lbs - with bearing lock applied. Pinch pins also had bearing lock applied and were torqued up to 10 ft-lbs. Campagnolo next time!

I went out on it today - an amazing feeling to be riding around on something you made - properly. My recumbants were always prototypes - hacked together. I gave it a thorough testing, over some evil jagged shale, at speed - nothing rattled. The whole bike feels and operates a lot more smoothly than when I got it from the bike shop - the cable routing has been rationalised, no weird corners now. I can tell the bike fits properly now - I can do wheelies!

Ridley #2

2009-02-19T01:43:00.000Z

Frame arrived today - I couldn't resist - I had to have a go at transferring the components - just to get a sense of what kind of bike its gonna be....I like it. It has a much better energy about it - plus its protected by the black, the red, the gold and the green...

It fits me much better, looks like no toe-overlap. Strangely the bars will end up just slightly closer - since the head tube is at an angle, as you raise the bars your going backwards too...

How anal is it to torque....?

GPS on a bike...

2009-02-12T20:04:00.000Z

Been Cycling with a GPS app running on my phone....

View Larger Map

...I was keen to have a go without shelling out for a phone, or contract - I've been a happy PAYG customer for a while now.

I tried out a few Java apps - native apps for Nokia series 40 phones seem to be a new thing, I'll be after a 6300i when they surface. I tried eGPSTrack, GSPJ, Mobile GMaps, TrailExplorer, TrekBuddy. The one I ended up using was TMJ-Mobile - it has a lot of features, and a bit of a learning curve - but its stable.

Nokia's Sports Tracker looks interesting, other services I looked at include MapMyRide, MyCyclingLog, Bikely, OpenStreetMap, MapMyTracks - the most useful I've found is MotionBased. That and a bit of Google Maps.

TMJ also provide some Java based map creation tools for your PC. With these I was able to pull the cycle route maps from OpenStreetMap and stick them on my mobile - this actually stopped us taking a wrong turn into an industrial estate...

Dirt

2009-01-21T14:32:00.000Z

I've just been out on my new Ridley - what a beast! When you get it into the mud, you realise you have two bikes in one...

osgOmniViewer

2008-12-08T00:26:00.001Z

Jim from The Elumenati and I managed to build a osgviewer based on their OmniMap library. These are by far the best results I've seen from OSG in a dome. It should also mean we can run a second node with frustum set up for our Zorro projector.

OSGOmniViewer from domejunky on Vimeo.

Make Human

2008-11-15T21:57:00.000Z

In researching my Foundation series animation, I've been looking at Make Human - this is a human imported into Blender and rendered with my fisheye Yafray renderer. I can't yet see how you'd make a lip-syncable mouth with it yet...

Long Hiatus...

2008-05-25T21:06:00.002+01:00

...Its been a while...

Its been a while, over a year. I'm going to update this blog with the stuff I've missed - but it won't be in any particular order.

One of the most recent issues is the search for a workable fisheye renderer for Blender. I've been experimenting with RenderMan renderers, which I'll document elsewhere. The issue here is with a renderer called Sunflow. This is a java based renderer that we could plug into the University's JGrid - which could be a few thousand machines.

This is the result from Sunflow:

Compare this to what we get from Yafray (which I modified and will document later)

I have a feeling that we're looking a different fisheye projections.

Really Slick Dome

2007-05-14T06:58:00.000+01:00

...Omnimap BETA...

The guys at Elumenati have released a beta of their Omnimap dome-correction library. This is quite novel as its the first post-shader dome-correction libraries - you can see my failed attempts at writing a vertex shader for dome on this blog. Omnimap is Windows only at the moment, so I used it as an opportunity to better acquaint myself with Visual Studio. This is a library designed to be inserted a the OpenGL level - so I decided to see if I could dome-correct one of the Really Slick screensavers.

The process was remarkably easy. I don't know C++ or Visual Studio or OpenGL that well - but I managed to find the 'draw' function, and add the relevant calls to Omnimap.

There are issues with licensing, I'm not sure of the license of the RSS savers - I'm informed that I need the original software to be LGPL or closed source for me to proceed legally...otherwise I'd have a copy of the screensaver here.

This is particularly gratifying for me. I have a passion for non-photoreal minimalist 3D. This seems like a natural extension of the minimalist movement - in that movement can be used instead of connotation. Dome-correcting 'Solar Winds' was one of the first bits of code I tried to dome-correct - I Intended to use this style of 3D as a basis for my 'Visceral Soup' for the dome.

Can't wait to see it on a dome - all that movement should be translated into movement of the viewer, definitely one to try on an empty stomach...

Moving Home

2007-04-10T22:44:00.000+01:00

...testing a new server

The box behind my server, is getting busier and busier, so I've rented a VPS/VDS from JVDS

So far so good - it is a FreeBSD 4.11 Jail - 4.11 is getting a bit long in the tooth - I don't really have the server space to have a proper ports tree going - and getting the right binary packages is proving troublesome...

setting the PACKAGESITE to this helps:

setenv PACKAGESITE ftp://ftp1.am.freebsd.org/pub/FreeBSD/ports/i386/packages-4.11-release/Latest/

as the standard php4 package doesn't the mod_php - which pretty much makes it impossible to make a binary only 4.11 FAMP setup.

[edit]

I passed on my comments to JVDS, and guess what? They do FreeBSD 6.1 too, great! So in about 30 minutes I achieved what I failed to achieve in 24 hours with 4.11...

Dome Flickr

2007-03-19T21:17:00.001Z

...getting Flickr images in the dome...

I found a Flickr 'group' dedicated to Equirectangular images. There is also a 'panorama' group, but this contains all sorts of images, but they are not equirectangular. So this weekend, I wrote a Flickr client for this group - enabling me to move inside these images, and render the output for the dome. Probably easier to show an image...

This is what is actually happening...

USB Pre-cognition

2007-02-16T08:36:00.000Z

I'm in the process of researching the kit for Mike Punt's 'Life Shirt' project. This is loosely connected to Dick Bierman's Pre-cognition experiments. We've had the Dome's seats fitted with USB sockets - but it was looking like there weren't many USB based bio-measurement products on the market. Then I came across USB-DUX

I spoke to two really nice guys, the company is basically a spin-off from Sterling Uni - Bernd Porr (fantastic name, makes me think of Steele Pulse for some reason) - seems like a fantastic guy! He wrote the driver and designed the circuit board. The horses mouth incarnate.

Bernd said he was unaware of any research where people had monitored more than 10 people - he thought the figure might even be 6. At present he thinks the COMEDI driver will only allow 32 devices, so he's up for the challenge of modifying his driver, and being able to say that his kit supports 40 punters. He said that he's be prepared to build up some of the pre-amplifiers for us, so that we could measure EEG. He was also happy to get one of his postgrads to solder up some sensors for us - not all that we'd need, but enough of each type to get us going.

All in all I think they might be the ones for us. Its cheap, and it makes us learn the bits of the process we want to become expert in, and outsources the bits that would be too much ball ache....

Omnidirectional Robot Vision

2007-02-16T08:26:00.000Z

...fisheyes and robots...

While looking for a hemispherical mirror, to play with Paul Bourke's technique, I came across all kinds of research using hemispherical mirrors or fisheyes for computer vision applications.

Dr Libor Spacek has video of a robot, based on a Gumstix board, avoiding dark objects.

I told Guido it would work!

Machine Vision Camera

2007-02-16T08:22:00.000Z

...the search for dome video continues...

I found some interesting links this evening:

http://dvinfo.net/conf/showthread.php?t=28781&page=33

http://www.cinematography.com/forum2004/index.php?showtopic=19400&st=60

Two threads talking about using machine vision cameras for cinema. This is basically what SI are doing, and it looks t me like they have better more 'camera' like control software.

The 'Pike' looks interesting from AVT, meant to be priced around $5000 - but it has a 1" Sensor.

http://www.alliedvisiontec.com/produktinfos.html?t=produktinfos&o=58&a=selectid

I downloaded some video from http://www.arecontvision.com - they have a couple of 180º/360º cameras, but also the AV2100 which is meant to be under $1000 - the footage is remarkably good - definitely better than a Sony Z1

Artistic Video Manipulation

2007-01-29T08:15:00.001Z

While looking around at 'slit' cameras, for panoramic uses I found an amazing collection of 'slit' camera art. I think a more creative side to my PhD is emerging. Feeding more from early cinematography, and the vocabulary you have to leave behind when shooting with a fisheye. But also embracing a more creative approach to visualisation. Like the following:

These guys have made a beautiful application for interactively navigating different sections of a timelapse.

This chap has created a camera from a image scanner, that seems to be producing equi-rectangular images natively. These are in the 100s of mega pixel range, they'd make perfect textures for my panorama viewing app...

He had a lens with 140º FOV - so in theory a 16mm Zenitar would do the trick (corner to corner). I read about Clifford Ross and his 'bad ass' 360º camera - I had to write to him when I read that there was no way to display his footage yet...

Adapting an Adapter

2007-01-24T01:15:00.001Z

...in camera truncation for the Sony Z1...

To try and squeeze a few more pixels out of the Z1, we have had an adapter made to shift the lens forward and up. Adrian - from Brunel Engineering labs did a fantastic job creating the adapter - I wish Sony worked to the same tolerances!

This is what the adapter looks like:

Looking at the most recent footage from the camera, I still think the HDV CODEC stinks. I can't believe that people all over the planet are signing its praises. Today, in a review of the Sony V1, I read that the BBC are using it. Maybe FullDome asks too much of it, but I don't think I'm alone - Panasonic obviously don't think much of it either - preferring to drop tape altogether for their HD camera and to go with their P2 solid state cards, in order to get enough bandwidth for a decent CODEC (DVCProHD). We were so ill-advised to buy the Sony!

I noticed that the Sony has a component video output - I went searching for a way to capture that direct-to-disk - in an attempt to bypass the Tape/CODEC. I'm not the first. Even capturing SD would be better than the results from HDV.

Cheek-to-Cheek

2006-11-30T10:26:00.002Z

...our Coastal Optics lens arrives...

I finally got a Nikon body to test the coastal optics lens out. Its been sitting in a box for a week, along with the rotator and iPix software. Its a beautiful bit of glass. When I attached it to the Nikon though, a D1x, there was a slight bit of give. This didn't worry me a first, but it turns out its just enough to crop the fisheye circle at the bottom.

I then installed the iPix software. I tried on my Intel Mac first, it wouldn't let me install the software for the dongle...OK, I tried on my Windows machine, the dongle software installed, but iPix software wouldn't recognise the dongle. I'd been wary of this software from the start. About a week after we ordered it, iPix went bankrupt. I though that we'd bought a dud. The next day I phoned our supplier, he walked me through getting the software going on Windows - turns out they hadn't sent us a license. They also informed me that the iPix software was known not to work on Intel Macs. Once he'd sent the licenses everything started working on Windows. With my new understanding of the dongle/iPix relationship, I tried installing the license files on my Mac, and what do you know, it worked!

The lens wasn't so successful, I had to make a short video of the wiggle, which has been forwarded to Coastal Optics...the lens uses a t-mount - which I've heard good and bad things about. I think maybe we just have to look out for one with a better fit. In the meantime, a bit of Rizla packet between the lens and camera does the trick.

I went out this evening - bored of making panoramas of my toilet:

This is what the combination of lens and software can do - this is in equirectangular format, which I can load into Panodome or QuartzComposer

I'm glad I got the software going. In moment of panic, I went and tried RealViz Stitcher Unlimited. Which since the demise of iPix has gained a back-to-back fisheye feature. The results aren't quite as good - it doesn't seem to do the colour matching that the iPix software does. Hopefully this will open things up for an OpenSource solution soon, iPix defended their patents quite fiercely - that was the reason for the < 160º limit in Panotools.

Ladybug

2006-11-17T21:58:00.001Z

...getting fisheyes out of Ladybug2 footage

I just phoned up Point Grey to get their prices, spoke to a nice women in Hamburg, unfortunately Point Grey are in Munich, so I redialed, they are nice people too.

I think we might have to discount the point grey as an option. I've been playing with their software. It would be easy to create a Quartz Composer application to playback the footage as it comes from the camera. If we want to get it into a fisheye form for editing as video things get a lot more complicated:

First you have to use their software to re-orientate the cameras to get an equirectangular like this: Otherwise we can't extract our 25º fisheye...

Then you move the image to a Linux box and use Gimp + the verison of Pantools that I hacked to do > 180º re-mapping on a 1:1 square that you select from the image.

Then rotate, it a few degrees

then crop it for 135º truncation

And that's just for one frame, this has to be done for every frame of the video! Certain parts of this process can be batched, but there has to be an easier way. That's a lot of processing, and a lot of errors to be added, the images aren't that good, or high rez to start with.

GML...

2006-11-16T18:49:00.000Z

...Object Orientated GIS?

I had a very stimulating conversation with an Archaeological Geographer on Monday. He has a very pragmatic understanding of GIS, and as a result has ended up teaching GIS to a lot of undergrads and masters students. He's the guy I've been looking for since I arrived here, and is happy to give me a bit of one-to-one to save me having to learn the whole field. During the conversation he mentioned GML - a new one on me, but completely logical I suppose. Raster formats are on their way out, and XML descriptions are on their way in. There is still room for raster data, but now it becomes a media type described, along with meaningful relational metadata, in the GML. One of the most interesting aspects of GML is the potential to describe temporal data, meaning walkthroughs of areas at different times should be possible.

This guy does field trips to Western Ireland, taking students on bus trips around the local area. His work in geographical archaeology looks at pollen in core samples, and extrapolates land use in the past. This should be an interesting set of visualisations for the dome. One interesting point he made was the most 'fly through' type visualisations are pretty meaningless, and that a 'walk through', ie from ground level is far more useful to him. He's looking to buy some GIS data from the Irish Ordinance Survey - from a brief bit of research it looks like OS are adopting GML in a big way. They even have some slideshows showing various types of geographical land features and how they are described in GML.

In terms of getting GML into the dome: It looks like GDAL has some support for GML (up to version 2.0) built in already. I've been using OpenSceneGraph with GDAL compiled in to get ESRI shape data in. So the tools are in place already. The bottom line is that since its XML based, you don't have to write a parser from the ground up. Something like libXML should give you all the tools you'd need, with some high level lookup tables to extract the features your after. That still leaves the 3D visualisation, but that's the PhD...

I was also pointed in the direction of UCL's CASA group, who are doing interesting stuff with GIS in gaming engines, we need to get these guys into the dome when its finished...

FullDome Video

2006-11-16T17:48:00.000Z

The options for capturing video for the dome

I've just spent two weeks wrestling with a Sony Z1E - trying to get decent video for the dome. Having jury-rigged a Century Optics onto it, in such a manner as to truncate the image for 135º 'in camera' - in an attempt so make the best use of the available pixels. The lens is only producing about 160º , and has trouble focusing to infinity when used in such a manner. Added to this is the HDV codec. I suppose when getting HD video onto the same medium as DV at 25Mb/s something has to give, and the noise of the resultant video is atrocious. Due to the 'long GOP' approach it is also very time consuming to edit and encode this video - this can be mitigated by using an intermediary codec such as the ones supplied by CineForm.

After a lot of cleaning and tweaking this is about the best it can produce

So the search is on for a replacement. I'm looking at cameras from Red, Immersive Media, Arri, Thomson, Point Grey and Silicon Imaging. Many out of our price range. Many it would be impossible to find fisheye lenses for. At the moment, the lead contender is the SI 2K Mini from Silicon Imaging.

I've been discussing lenses and things with their engineers, and checking out their workflow. They have teamed up with a firm called CineForm, who I discovered though a set of intermediate codecs to make the Z1 more palatable. They've developed a 10 bit 4:4:4 video codec called CineForm RAW, which give a 5:1 compression, and Silicon Imaging are employing this in their cameras. I'm stunned by the results, here's a few links you might be interested in:

This is the firm that did the Citroen adverts, they're doing a feature film using the SI cameras - pretty good insight into the camera:

This is SI's main site with links to some light details on the codec:

Here's the griff on the codec:

This cameras is out in the wild, and the 2K Mini is only around £6500 - which is basically a head that you attach to a NAS or a decent laptop via GB Ethernet, and run their control software...and get editing...pretty sweet. It'll take these lenses:

Coastal Optics, Fujinon

For us its a pretty workable solution, the combination of price, easy workflow and lenses makes it the only horse in town...

The Janner

2006-11-05T05:55:00.000Z

my variation on a classic cocktail

I was watching The Big Lebowski recently - one of my favorite films - I watched it twice this week. During the film he drinks 'White Russians' constantly. In a few scenes he refers to them as 'Caucasians'. Well I've developed a variation:

1 part Gin (preferably Plymouth Gin)

1 part Kahlua

2 parts Milk (or 1 part Cream)

Serve with plenty of ice, and you have yourself a 'Janner' - the local term for someone born and bred in Plymouth.

Windows...After all these years

2006-11-04T06:23:00.000Z

...not quite an old friend

Having abandoned windows years ago for Linux, then FreeBSD, and latterly a combination of Linux/FreeBSD and Mac OS X...I've found peace in Windows XP. Putting together work flows for dome content will push any hardware software combination to the max. I got tired of having to learn, or trying to find 3 ways (Mac/PC/Linux) of doing every task. From 3D modeling, to video editing and encoding at extreme resolutions to real-time 3D - much of what is contained on this site.

The hardest test seems to be getting full resolution fulldome video to playback from a single hard drive - full resolution here means anything from 1400x1050 or 2048x1536 right up to 4kx4k. Speaking to the guys from Elumenati at a recent conference in Portugal - they seemed to be getting amazing results from Windows Media 9. I'd been using Final Cut for editing, and was having trouble finding an appropriate deployment codec. Even resorting to Apple Intermediate Codec, to try and keep the quality. At this resolution
its tough getting good performance out of H.264. I have a loathing for Adobe Premiere, so didn't really want to start using XP as my main platform. Enter Sony Vegas. Although it looks like a toy, its actually a solid application - the bonus being that it does 5.1 audio too and its about a third of the price of Premiere. Encoding to WM9 does indeed produce amazing results. I've been following codec developments at Apple for nearly 5 years - I wonder now if behind all the rhetoric of standards compliance, is the
fact that they have to be - people have been burned by the Apple proprietary cul-de-sac before. Microsoft don't care - they'll throw money at codec development until it does what they want, and if that means it only works on Windows - so much the better. It's good to find a solution, and to start concentrating on content. I should have a bit more to write here now, rather than digging through the manual to 'mplayer' wondering how to crop video in YUV space...or whatever...

OpenSceneGraph works here too (of course) Panda3D only seems to work here, SEOS' software is Windows only, Uniview (SCISS) is Windows only. I'll miss the attention to HCI in the interface, and the general pleasant-ness of Mac OS X, but its the content that matters...Never thought I'd be saying that about Windows XP!

I still can't find a substitute for MOTU's Digital Performer - nothing I've ever seen or used touches that for music and 5.1 production, but I can live with that.

Back-To-Back

2006-07-14T22:18:00.000+01:00

Getting back-to-back fisheyes into the dome...

I've found a methodology for getting back-to-back fisheyes into a realtime dome-corrected panorama viewer....

The software I've been using is a combination of GIMP ( http://gimp.org) and Hugin (http://hugin.sourceforge.net/) which use a modified (by me) version of panotools (http://panotools.sourceforge.net/)

I took 2 back-to-back fisheyes taken with the Coastal lens:

made equirectangular views of them using The GIMP:

Which will stitch together quite nicely - Hugin does an even better job of this - but I just did it by eye for now which produced this:

Then in Quartz Composer I created realtime visualisation by wrapping the above around a sphere:

The finished article:

Dome OpenSceneGraph

2006-06-20T22:14:00.000+01:00

I've abandoned vertex shaders for a while, and have hacked one of the OpenSceneGraph examples to do dome correction. This uses a 6 camera frustum looking at the scene. The output of which is then stitched and mapped onto a sphere as a texture. In our case we need the view from the inside of the sphere rather than the external surface.

A vertex shader is being used to the texture mapping - and on a 1.83Ghz Macbook Pro it can manage 60fps with a texture size of 512.

This is quite a moment: I have been trying to achieve this in my spare time since September 05. This means we have a standard dome-corrected platform for 3D models, Geospatial data, streamed video, Meteorological data. I've used large APIs before (namely QuickTime) - OpenSceneGraph is actually a pleasure to use. It has high level features that mean you can get things working quickly - but doesn't punish you later when you need to drill down for some reason.

Surround Sound

2006-06-01T19:38:00.000+01:00

getting the rainforest into the dome...

We have a requirement for sound capture for field trips. One academic has suggested/requested capturing surround sound.

I have been fiddling with surround sound technologies for years. Mainly encoding to Dolby Pro Logic - which is a method of encoding 4 channels of audio into a stereo pair. This technology has been around for years, most films broadcast on public television still have their pro-logic encoding intact. Even some soaps and dramas (Eastenders, ER) are broadcast with Pro Logic.

I bought myself a Dolby Digital amplifier for my birthday. Dolby Digital is the next generation from Pro Logic. Dolby Digital encodes various numbers of channels, from 2 to 7.1 into discrete audio tracks, which are then multiplexed and sent down an S/PDIF connection to a decoder. Most modern Macs, and PC soundcards come with S/PDIF connectors nowadays. When playing DVDs or AC3 files, the encoded Dolby Digital information is sent direct to the decoder. There is no technology that exists as yet (other than high-end hardware built by Dolby Labs) which can encode to Dolby Digital in real-time. So to monitor and mix Dolby Digital soundtracks you need a multi-channel soundcard, and either discrete speakers, or discrete inputs on you decoder.

In my experiments, I have only ever played with synthesising surround sound, the challenge here is to find a microphone technique for capturing the surround information. The most interesting avenue for researching this is the world of classical music TV broadcast. Across Europe many broadcasting companies are readying themselves for the time when Cable and Satellite can deliver surround information digitally. The Austrian and Danish broadcasting companies in particular, have pioneered microphone techniques that have challenged the audiophile attitudes toward listening to classical music recorded in surround.

Current state (techniques)

Off the shelf (Schoeps)

Mid/Side - double stereo

Recording interfaces and 4 track machines

DPA Mic Techniques

Schoeps
Jerry Bruck on KFM 360
Claes' Mic Tests

Cheap Ambisonic setup

Jungle Fever

2006-05-31T00:11:00.000+01:00

Fisheye lenses in the field

We have a guy out in the jungle in Borneo at present. He's taken the Nikon 8800 + FC-E9 fisheye converter. This is the second trip to the jungle for this camera. Previously it was sent with the University's photographer. On occasion he'd get a 'System Error 2' - this seems to be a common fault with these Nikons - and was coupled with the fact that the camera was being carried by a student in between shots, and would come back with a random set of setting each time. I had figured that it might have been an isolated incident. But our man in Borneo has just reported having the same error. Its a shame as he was producing nice shots, and we'd sent him out with a 4GB CF Card - enough to get a day's timelapse.

Primary rainforest must be one of the toughest environments to take a camera, I suspect that moisture and humidity are to blame. I suspect that when the camera comes back it will work fine - as soon as it dries out. I does mean that we need to consider a new camera. This was on the cards anyway, as the quality of the FC-E9 circular fisheye's aren't great:

Even with this amount of compression, there is clear chromatic aberration, and a strong blue bloom around the edge. These are artefacts common to many fisheye lenses, but are particularly pronounced on this lens. My partner also has this camera and lens, and is displays the same behaviour - as do the many KAP photographs taken with this combination.

Lenses

Finding a circular fisheye for a digital SLR isn't easy. Fisheyes are rare/specialist lenses anyway, the 'crop' factor of most current digital SLRs renders most 7-8mm lenses useless. Since the CCD in these cameras is smaller than traditional film, only part of the circle is captured. There is in fact only one lens which can be used by 1.6x or 1.3x digital SLRs, this includes all of the current Nikons and Canons except a few models I'll mention in a minute. This lens is made by Coastal Optics in the US. According to the fisheye bible, this is the same lens as marketed by iPix.

Canon make two cameras that have full-sized CCDs (actually CMOS) - the EOS 5D and EOS 1Ds MkII. These are 12.8 and 16.7 MP (Mega Pixels) respectively. The list of fisheyes available for these is slightly larger:

Sigma - We bought one of these in Nikon fit by mistake. Don't let and of the "For Digital", "DG" or "EX" confuse you. It is only a circular fisheye on a 36mm x 24mm image plane.

Peleng - This is a Russian built lens. It seems to be hit and miss in terms of quality. Hertblei, a Czech firm who rebuild Russian and Ukrainian cameras and lenses are said to produce a multi-coated version, but it doesn't appear on their list.

Old Nikons - These come up from time to time and can be adapted for Canon EOS mount. Ironically, early FD mount lenses for pre-EOS Canons, don't do so well.

Coastal Optics - As well as the lens mentioned above, they also produce one designed for 35mm film cameras - Coastal Optics confirm that this would fit a 5D or 1Ds. Interesting to note is that the digital version has f-Theta distortion < 6% whereas the 35mm version is < 2%. This lens is also double the price of the digital version.

Given the above list. The Sigma and the Peleng are the cheap options - both are available for less than £500. The Nikons are available occasionally for 1000s and the Coastal Optics is $7500. The next step was to evaluate the Sigma vs the Peleng.

There are some technical papers available on the Coastal Optics site. This chap has a good overview of the Peleng. This site has some tests for the Sigma. The real evaluation needed lots of images from each lens, below are the best from those I found on the internet:

Peleng

Sigma

PBase.com is an amazing resource for finding images like the above. They're searchable by camera, lens or keywords. The above are a mixture of Canon 5D and 1Ds MKII.

There doesn't seem to be that much between the lenses. They still show the blue ring of the Nikon FC-E9 - this is quite soft and wide with the Peleng, thinner and more pronounced with the Sigma - the Sigma would be easier to remove in post-production. They both also display signs of chromatic aberration. The Peleng is only slightly better than the Nikon FC-E9 - the Sigma is much better here. The final judgement is subjective, an authoritative assessment of sharpness cannot be made given various processing and compression these images have undergone. The Sigma though does seem to be sharper and more 'contrasty'

Bodies

There is a lot of discussion on photography mailing lists about weatherproofing. Some have the opinion that cameras are usually more weatherproof than their owners - others (including the Manufacturers) who go to great lengths and costs to make their kit weatherproof.

The EOS 1 range, is Canon's premier professional range - and as such has features designed for expedition use. All sockets, buttons etc are weatherproof. Used with Canon 'L' series lenses - it is said to be tropical proof. The 5D is part of Canon's 'pro-sumer' range, and as such they make no claims about durability under harsh conditions.

There is a huge difference in price - you can almost buy three 5D bodies for the price of one 1D. It has a higher resolution sensor, and many features to ease the life of a pro-photographer. Most of these would be wasted on us. The only thing that helps me with this decision is thinking that a 5D which has been flown around the world, and ends up failing in the forest, is not cheap at all...

Vertex Shaders

2006-03-27T18:40:00.000+01:00

hmmmm....

I've been playing with OpenGL vertex shaders recently. Trying to get a more efficient method for real-time dome correction.

This is the clue I was given...care of Mike Bailey:

http://web.engr.oregonstate.edu/~mjb/WebMjb/Projects/dome.pdf

I have this running in Apple's OpenGL Shader Builder...sort of. It definately deforms my shapes, but I'm not sure if they're dome-corrected yet.

I have also using this with OpenSceneGraph. We have a model of the campus that has been saved from 3D Studio Max via an OSG export. This is then loaded into an OSG app I've created which loads the vertex shader:

There are issues with texture mapping and culling, but it's a start...This is an image of our dome, with a tree behind it, they're not flames.

Schedule (Gantt Chart)

2006-03-14T07:29:00.000Z

Gantt Chart

A Tri-Cornered Hat...

2006-03-09T20:26:00.000Z

Exploring the tetrahedral method of stitching dome masters

I've been playing with the Tristitch or tetrahedral method of creating dome masters

I set this scene up in Blender (3D package) - that mess in the middle, is 3 cameras setup like the pineapple site...

Pineapple

I rendered the view from each camera

Front:

Right:

Left:

This is the output from Tristitcher

This is what it looks like if you leave the two rear images off

This was done to test the quality of the stitching, I figured aligning text was about as hard as it gets...next I'll render a few 'scenes' to see the effect of leaving the rear cameras out. I haven't used real cameras yet, as I don't have a method of working out FOV for any given lens particularly mid-zoom. And it turns out this is critical. I think the above is promising - its far quicker and less ungainly than the cubic method - if only I could find the maths for it, it would transform realtime rendering dramatically...

I made an animation...all low resolution. One thing I went and checked on, that pineapple site has loads of software for sale, tristitcher seems to be the only product which is free....spot on!

We can now reliably create pre-rendered 3D using free tools. In a form that can be queued up for rendering on a render farm

Asterisk*

2006-02-13T03:00:00.000Z

Getting Asterisk* working with a Speedtouch 190 ATA

I've just resurrected my Asterisk* server. This time I'm running it in a jail, I copied over my old conf files, and most things worked. I never really got my speedtouch 190 going before. This is a little box that has an ethernet port to connect with your LAN, and a phone point for your ordinary phone. It cost me £14 - with £10 of call credit - so a lot cheaper than a hardware SIP phone.

The sticking point was getting Asterisk* to recognise DTMF tones. It worked fine to call an outside line, or to call voicemail - but once a call had started, no DTMF tones would be recognised. The trick is to use RFC 2833. For Asterisk* I had to change sip.conf. Change the line dtmfmode to read:

dtmfmode=rfc2833

For your Speedtouch navigate to the SIP section, and look for the "OOB Signalling" page. Chnage the settings so that they look like this:

FreeBSD for Video

2006-02-04T01:12:00.000Z

Using FreeBSD for Video and DVDs

I've just spent the last week using FreeBSD to create some DVDs for my Brother's birthday. I am a fan of a UK series from the early 90s called "The Secret Life of Machines". I downloaded three series from the web encoded with Xvid mpeg4. At first I tried using Sony's DVD Architect. Its a nice bit of software, after downloading an array of kiddy-ware and codec packs for Windows - in an attempt to get DVD Architect to load the Xvid files without conversion, I stumbled across the fact that you can just change the four char code to DIVX/divx and both Vegas and DVD Architect will recognise the files...or so you think. It turns out that DVD Architect doesn't like variable bitrate mp3 - so the resultant DVDs have skipping audio and considerable lag between audio and video. A bit of VirtualDub meant that I could extract the audio to uncompressed PCM, and add it in as the soundtrack in DVD Architect. Some of these lessons are hard to learn when the processes take so long, re-encoding the video and burning the DVD took about 9.5 hours on my 2Ghz Athlon XP.

This all worked great for 2 of the series. The third kept failing after about 6.5 hours - no errors, it would just stop updating the progress bar. The quality of the software I was using (VirtualDUB, codec packs etc) was giving me cause for concern, and most is pretty ugly anyway, so I decided to get to grips with doing it under FreeBSD. I've been pleasantly surprised lately how good Mplayer is when you do more than scratch the surface. I don't use the interface, the power is hidden in the command line. I won't watch TV/DVDs without Xvidix extensions running nowadays, and figured there must be a way of re-encoding DivX to MPEG2.

I ended up following this Gentoo how to. If you know FreeBSD, its pretty easy to adapt. Rather than writing my own XML for DVDAuthor, I used DVDStyler. To get DVDStyler working properly, you need to add support for ATAPI to your kernel:

options atapicam

Then add the apprpriate links and permissions in /etc/devfs.conf - mine looks like this (the pass0 was a new one on me)

link acd0 cdrom
link cd0 dvd
perm acd0 0660
perm pass0 0660
perm cd0 0660

I had to modify the transcode flags to compress the MPEG2 to get six episodes onto one disk and to maintain the 48khz audio - I was getting squeaky soundtracks before this. and make sense of the 4:3 aspect of the original. I ended up with the line:

transcode -i Secret\ Life\ Of\ Machines\ 101\ The\ Vacuum\ Cleaner.avi -y ffmpeg --export_prof dvd-pal --import_asr 2.21:1 --export_asr 2 -o Vacuum -D0 -b224 -N 0x2000 -w 3000 -s2 -m Vacuum.ac3 -J modfps=clonetype=3 --export_fps 25

Once you've set this up you should be able to do use the following to 'mux' the audio and video into one MPEG2 file.

mplex -f 8 -o matrix_dvd.mpg matrix.m2v matrix.ac3

Which DVDStyler will accept as compliant content, and away you go...The last thing which bit my bum, was setting the Menu frame count in DVDStyler - I used a figure of 180 - without this your DVD player will flash your menus at you tantalisingly for only 1 second...

November Dome Update

2005-11-10T21:40:00.000Z

...its been a while

I haven't posted for a while. I have to keep this blog a commercially-sensitive-information-free zone. At least until the tendering process for our dome is finished.

Quartz Composer vs Panoramas

It looks panoramic images will be important to the dome - I've been looking at various ways of creating them. iPix and QuickTimeVR seem to be the main methods for playback. The problem is that they produce an equiretangular view - really we need the underlying panorama - or a method to render out a cubic QuickTimeVR to a dome master. Given the original spherical equiretangular panorama, Quartz can produce stunning results. I've done some tests with the output from a LadyBug2 - both moving images and stills - these work well too. iPix gave us a link to one of their files - its big. When wrapped to a sphere in Quartz Composer - the results can be panned vertically and horizontally, with no visible seems.

SGI

I've been given an SGI O2 by the neuroscience department. Its a bit long in the tooth, but it should serve its purpose in getting me up to speed with IRIX. I use FreeBSD a lot, and have used NetBSD, OpenBSD, Solaris, Debian, Gentoo, Red Hat...so I feel I have a pretty good overview of what Unix looks like. IRIX is hard. The installation is cryptic, there's no package management - but then the system wasn't designed for pragmatic sys admin...

The main reason for playing with a 5 year old O2 is to check out the 'famous' SGI libraries. Many of which have been replaced by open source equivalents, or have been given to the open source community. Its easy to see why SGI dominated the visualisation arena for so many years - Performer is a powerful library.

Mac/PC specs

Getting off-the-shelf Macs and PCs that will run and create dome content is quite a challenge. The weak links are the hard-drives and the graphics cards. The latest generation of cards from nVidia are capable. The Quadro range (FX 3400, 3450, 4400 and 4500) and the GeForce 7800 GT/GTX will do the trick - but they are expensive cards. In terms of disk speed, RAID is the way to go. I have just setup a FreeBSD 6.0 machine with RAIDed drives. Very practical under FreeBSD.

I have some concerns over speccing Windows machines for the project. Apart from some users having a requirement for 3D Studio Max - there is little requirement for it. In terms of real-time modelling, I would hate to write something in OpenGL then have Microsoft pull OpenGL from Vista. Mac OS X looks like a safer bet, but the threading and IO issues worry me.

I need to keep an open mind on this. Just as Quartz Composer on a Mac would enable a lot of dome content - easily, there might be environments in Windows that would enable the same. I haven't found them yet, but they might be there. There is also the fact the most of the Planetarium equipment suppliers use Windows machines for the playback devices.

The compromise I've come to, is to find a spec that will allow Linux/FreeBSD and Windows. That way the box won't end up being used just for 3D Studio Max. The trick here in my experience, is to buy a machine based on an nVidia graphics chip. Linux/FreeBSD can be made to work with ATI graphics cards through DRI (part of X11) but it never seems to get the full speed of the card. nVidia, on the other hand have produced native binaries for Linux and FreeBSD for years. Some complain that they're not open source - but there is a lot of IPR in the GFX business...

An ideal solution would be to find an nVidia based laptop that would run FreeBSD. These are quite rare. I've found a discontinued model on Ebuyer - a sempron based Acer with a GeForce FX 5200 - having owned an AMD based laptop before, I worry about heat/noise/battery with this. The other option I've found is a Dell - a precision M70.

Week Two...

2005-09-30T23:55:00.000+01:00

...More experimentation

Mike and his Jesuits...

Mike punt walked into the office the other morning, and remarked on the similarity between my dome projections and anamorphic projections. He was looking at a photograph I'd taken of a fisheye-like tap in my bathroom:

Meeting with Allison

I had an interesting meeting with Allison Stokes from the CETL. It became apparent at this meeting that there is some confusion between "high production values" and "immersive theatre". It will be interesting to see if some of the claims made for domes hold up when compared to high quality 2D visualisations .

Experimentation...

I've been continuing by exploration of free tools for dome projection. I've found a few:

Stellarium

This is a free (OpenSource?) Astronomy program, designed for dome projection. Shown here is the end of the partial eclipse on 3rd October. I was looking for a free alternative for Uniview from SCISS. I transpires that Stellarium and Uniview are not the same animal at all. Stellarium is geo-centric - Uniview can show you the view from any point within your dataset. It is also being moved towards a state where it can take many differing types of dataset, so it is not limited to astronomical data...

Quartz Composer:

Quartz Composer is a free tool included with the Apple Developer Tools. It is geared towards graphics rendering, but is actually a 5th generation programming environment. It includes components for live video capture, video streaming. David McConville of The Elumenati talked of using Quartz "...cos its there..." - but I think it has more potential than that. At present I am using a spherical renderer, and applying a transform to the rendered image before I wrap it to the sphere. Of course in any other environment the previous sentence would have meant hours of hard coding - but in actuality it was a few clicks of a mouse...
Domes:

Its becoming clear that a "development dome" will be a necessity. I have been looking at various options:

This is "LarryDome" used by the people at LochNess - it's made by Shelter Systems and costs $620. The inflatable options are more expensive. There are some very interesting domes available from this UK manufacturer Stargazer:

...The one pictured here is constructed like a drum - the bottom skin is then 'sucked' into the shape
of a hemisphere. They allow for 360 degree access, as punters can just walk underneath the dome.

...These domes are made by a guy from Barcelona I met in Portugal, they cost in the region of €6000. I spent most of the week on Portugal sitting in one of these. I watched Harald Singer's Romeo & Juliet (sitting next to him) in one of these, and the results are convincing.

Notes From Espinho

2005-09-28T00:39:00.000+01:00

The rough notes from the 2005 Immersive Theatre Conference

Ed Lantz' Intro:

Immersion environments stimulate the 'opto-vestibulator response'.

Anaheim system uses 48fps which removes jitter. Evans and Sutherland use 60fps for real-time, 30 fps for pre-rendered.

Typical tilt for domes is 10-30 degrees - usually worked out using sight line. (i.e. the heads of neighbours) 20 degrees is common.

Differing models for rendering:

Cylindrical Map: 1 pixel at top
Equidistant Polar: (de facto standard) 1 pixel at equator

Platonic: Various shapes, most common is Cubic - often used for real-time

Fisheye

(much simpler workflow, limited by resolution of projectors)

Elumens VisionStation, and Spitz SciDome are small scale systems possibly suitable for preview system.

Multi Channel

Onyx (or similar) pipe per projector, 5,6,7 up to 41....recently can use high-end PC. Maintenance is a big problem. DLP projectors require optical mask due to lack of 'real' black. Active blending systems exist using cameras, but some doubt that they work (Barco)

2D Image remapping, often called 'Billboarding', suitable for re-use of 2D assets. Any interlaced content will need conversion to progressive (video)

4,2,2 and 4,2,1 lead to loss of colour - 4,4,4 better

Immersive Media - www.immersivemedia.com

Dalsa - www.dalsa.com

More links: www.cis.upenn.edu/~kostas/omni.html

Company Showcases:

Navegar: (hosts)

Wanted to learn how to produce content, then build system to fit. Produce AfterEffects plugin with Oporto Univeristy - will be going gold in October.

SEOS: Ian Dyer

Have done research into educational uses. Recommend 160x40 degrees for immersion.
SPIDER: Renders to .rgb files
SCORPION: For repurposing all media types, can project discrete channels of content over main projected image.

Prof. Nick Avis (Cardiff) - interesting guy for research

SYK-SKAN

Software: DigiDome (Slide manipulation); DigiSky (Astronomy); SPICE (content authoring - from SEOS)

Hardware: Definiti - complete portable solution - inflatable dome (€400,000 + dome €6000)

"Content is Still King"

RSA Cosmos

Produced screen for Queen Mary II. Content creation tool is Linux based, as is live control software. Huge variety of projector systems - very Astro-centric

Living Globe: Harald Singer

Produced Romeo & Juliet - handle on narrative from film and television. Plans to produce LED based 'pearl' in Dubai. Announced imminent availability of 4Kx4K camera.

"Directing [dome productions] halfway between film and theatre"

35mm ARRI Camera (Sweet!!) + 6mm Nikon fisheye. One of only two in existence, and uninsurable. Equates to 220 degrees.

Scanned at 4096x3112 to Cineon - OpenEXR format (ILM) (Yafray???)

Final resolution was 3120x3120 as 24bit targa files. Don't need warp or distort for dome projection. Took 2TB to build, 0.75TB to store, 4TB including backups etc. This is for 20 minutes of video.

Found that conversations across dome were too much, better to use 3/4 of dome.

Cinematographic Techniques:

Tim Hope - Hamburg Planetarium

Talked about Francois Truffaut's 'Auteur Theory' coming into play with domes. Tested projections at Hambug with 1K, 2K and 3.2K - results were 'no perceptible difference'.

Talked about Eisenstein, Hitchcock et al - Difficulty of using differing shots from world of cinema. Reminded me of Jaques Tati's "The close-up is vulgar" preferring prefering to let viewer look at their choice.

Used to work for allsky.de - good source of dome master stills. Thought concentric domes better for immersion. "Tilt means you get distracted by the dome" this was a minority view.

Mentioned Lev Kuleshov/Vsevolod Rudovkin and Alekmodr Solkerov - 2002 - zero edits/cuts - Tillman Buttner (Living Globe) Piece recorded using steadycam.

Mirage 3D

Simualation software and hardware. Cycling simulation, Driving simulations (how to save fuel)

"World Vision" - Real Time optimised database

"Origins of Life" - Photo-realism, so used Maya. DNA fly-through - in style of electron microscope (although this isn't possible) - AI for butterfly movement.

Directing Full Dome Video Show

Using thesis (degree) students as employees, takes more time and hand-holding, but worth it. Student's scenes must be indistinguishable from pro's.

Procedure (his)

Story board
Interviews
Music
Editing
Keyframing + Edit - dictated by music
Script comes last

Photo-realism using D-Sculpter. 40Ghz render farm was far from enough < 10 hours per frame.
Used local University's spare CPU for render farm.

Motivational visits for textues - ie went to Arian rocket.

Evans & Sutherland

Micheal Dout

Digistar 3 (same system as Thinktank in Birmingham. Ability in control software to have bespoke interface per user. Ability to show Video + 3D.

E&S were responsible for first full sphere dome

Interactivity is often social - via conversation

Provide on-site training and support if you buy their products - domes less than 10m usually don't need more than a week of support.

Just produced the SP2 (Fisheye?) capable of 1400x1400

Digistar 3 - Producer

$2500 for library access or 25% of gate takings - this gives access to 2 shows per year.

Scripts and Storyboarding:

Strengths and weaknesses - reference to McCluhan - surface area of dome far greater than even iMax.

Pros

immersion
TV Playing catchup
Dome finally matches 5.1

Cons

Only one lens. Can reproduce the effect of dome others. R&J shows difficulty of working with this
No live-action camera
One good seat
Audience sizes are small
Visible seems (not with Plaster of Paris)

"Hand them a textbook" - idea of "forced brevity"

"Show not tell"

Create 'Animatic' (animated time-accurate story board) edit to timing of scratch narration (...or music like Mirage3D?)

Remember "Process is not a checklist"

SKISS

Staffan Klosier

Swedish firm, come from strong gaming background. Produce Uniview - version 1.0 is out

ScaleGraph - no content yet
Accessibility
No sell - more interested in partnerships

http://www.laserium.com/cyberdome

Dance interactivity

Mentioned Unreal Tournament engine which is available for $20 (?) (Quake Engine GPLed?)

Mentioned coolpix + monopod guy from 'inhtour.com' (Spelling?)

"Machina" - video with realtime software

David MacConville

NooSpheric Research - The Elumenati - papers at fulldome.org

Stan Vanderbeek - dome experiments - media artist - MIT...networked dome

Subscribe to "Planetarium" - recent article about immersive video

Benefits of Virtual Environments:

Complex spatial concepts - "perceptualise"
Engagement
Differing spatial/temporal perspectives
Motivation
Better task performance
Higher satisfaction
Presence (Marvin Minsky) telepresence
Increased Attention
Increased pervasiveness
Improved reflection and memory
Potentially increase the effectiveness of any material

Why Domes?

Millennia looking at the dome of the sky. Carnival/Collaborative

"Communal ego-centric perspective"
"How do I see things" - can't zoom etc when filming/making things. Matching unaided eye.

Henri Poincare - "Language of New Media"

Realtime+Realspace
Temporal vs Spatial
Montage vs Compositing

Willheim Nestle "From myths to logos"
Ben..." Exploring the Frame"

Skiss - Part II - Uniview

Development experience of UniView

OpenGL Graphics should be on par with Playstation to compete with game culture.

"Halo Surfaces" - bump map/Normal map - Zero impact due to GPU

Specular Map (B&W map of reflectivity)
Ambient light (City lights)

Uniview not only planets. ScaleGraph, using local co-ordinate systems.

"Flight Assist" - Usability. "No Magic" - inertia, extensible via pretty easy OO language - (afterward talked of being let under the bonnet in return for Mac OS X interface)

Working relationship between David and Staffan - interested in help with admin tasks on fulldome.org

fulldome@yahoogroups.com

NVidia interested in dome functions on card. Talk to David.

SEOS - Ian Dyer

Contrast & Media Storage...Audio

Resolution - Preferred measurement is the Arc-minute - this is independent of resolution.

1 arc-minute = 100m pixels
3 arc-minutes = 10m pixels
6 arc-minutes = 6m pixels (common for large theatres)

Genlock/Snaplock/Framelock - important for multi-channel domes - NVidia are best at this.

Media Server

Realtime:
RAID (High end SCSI - SATA doesn't cut it)

60fps - 1280x1024

Dual Xeon 3Ghz

Stereoscopic

Screen gain of 2 means traditional technologies don't work in domes. SEOS have developed "Infotech" which add controllable grid (bit like CMYK?)

Navegar - AE Plugin

Cinema4D - Stitching - Rendered with 5 camera setup with 90 degrees FOV for each.

Up = 90 deg. Altitude, 90 deg Azimuth + Hard contour
Right = 0 deg. Altitude, 90 deg. Azimuth

Video - can stitch video or video array as easily as stills. Also create dome master from assets in usual AE manner. Mac compatible.

Micheal Dout - E&S

Audio for dome productions.

Invest in:

Acoustic treatment behind dome - perforations allow sound to hit insulation behind dome.

"Contact" - free software, act like sampler.

"Fletcher Munsen Curve"

Surround

Start with stereo, route to sides (ie music)
Send music to rear, add reverb and delay, 5-10ms to avoid phase cancellation.

"Domes Colour Sound!"

Digistar III has SMPTE so can be controlled by audio software.

Week One...

2005-09-23T23:24:00.000+01:00

My progress during the "comedown" phase

The Experiments...

So, I have set-up a dome to play with, using a £6 Ikea lamp. This works ok for what it is. It helped me understand the role of fisheye/wide angle lenses in the projection phase of domes. That is, they are not absolutely necessary. At present, the projector is about 1 1/2 metres from the dome. If this were scaled to a full size dome, then a 10m pit would need to be dug to accommodate the projector. Hence the need for wide-angle lenses. The other reason is the wide depth of field. With a small dome the distance between the projector and the edge of the dome vs the distance between the projector and zenith of the dome is negligible - but with a full size dome this becomes an issue - having a wide-angle/fisheye means the projector can keep the projected image in focus across the dome.

I have been doing some experimentation with Blender and Yafray. Hidden in the documentation for Yafray is a camera type of "lightprobe" - this produces a spherical image:

...Unfortunately, it turns out to be a 360 degree view, rather than 180, but its a start. I am now looking into using Blender with POVRay - which is know to produce 180 degree spherical output.

Following up on a post by David McConville on FullDome.org - I have been looking at "Fisheye Quake". This had gone a bit stale - it was a Windows only hack. I found the source code for the hack, and coupled with the fact that Quake 3 was GPL'd a few weeks ago, I'm confident that we can get this going in the dome. The interesting part of this is that "modding" applications for Quake 3 are quite common, potentially opening up the quake engine to Artistic/Visualisation activities...

Blogging Up!

2005-09-22T23:59:00.000+01:00

Well, I've got the blog software installed.

Blurb

2005-09-03T03:30:00.000+01:00

Introduction to the project:

Pete Carss is a Research Assistant in Immersive Vision at the Institute of Digital Art and Technology.

The Immersive Theatre is a transdisciplinary project born out of the Experiential learning in environmental and Natural Sciences - CETL. The aim of the project is to transform a defunct Planetarium into a state of the art immersive learning environment. A diverse range of disciplines hope to revolutionise their teaching practice and pedagogy, through the use of real-time and pre-rendered 3D simulation; 360º video and stills. In addition, digital practitioners; audio and visual artists will use this space to explore future applications of immersive spaces.