In the middle of the night one sometimes turns over to see a (boogey) man standing in the bedroom but it will be, upon further thought, just a coat hanging on a coat hanger.
When algorithmically determining what a machine will imagine is there in an image it is not normal to include boogey men in the list of possible things the machine might imagine. However it is always necessary for the machine to imagine something. It forms an object hypothesis. Anything in the image that is inconsistent with the object hypothesis is simply deemed as unknown.
The human brain does not appear to immediately relegate inconsistencys as unknowns. It will continue to imagines all sorts of things, whether it tests such for consistency or not. The last thing it will do is give up on imagining what is there in the image.
But my algorithms give up immediately. Where it can register what is there it will do so and but everything else is assigned to the set of unknown things in the image.
Monday, July 20, 2009
Thursday, July 16, 2009
Estimating specifications
When accurately estimating how long it would take me write a specification I always end up with the same estimate: zero time. This is because the more accurate the estimate the more specified must be that which I'm estimating. So when the estimate reaches complete accuracy the specification reaches complete completion (so to speak)
So there is nothing left to do.
To avoid this problem I now give inaccurate estimates.
So there is nothing left to do.
To avoid this problem I now give inaccurate estimates.
Wednesday, July 15, 2009
Video generated geometry
Creating digital geometry from video is an ongoing research project of mine. I'm currently writing design tools for a documentary film where animation is created from archive footage from the thirtys and forties. The algorithm uses motion vector analysis and camera angle reconstruction techniques to decompose the footage for subsequent restoration. The algorithms separate out rigid and non-rigid motion effectively separating figures from their background. The background becomes re-encoded as a 3d model which aids restoration.
Thursday, July 9, 2009
Actionscript Virtual Machine
Have been playing with the source code of Tamarin, the Actionscript virtual machine inside Adobe Flash Player. I put this intro together:
http://members.iinet.net.au/~carllooper/tamarin/HackingTamarinTracing.htm
I've now got my own custom virtual machine running in which I can call OpenGL 3D graphics hardware from Actionscript!
Love it.
http://members.iinet.net.au/~carllooper/tamarin/HackingTamarinTracing.htm
I've now got my own custom virtual machine running in which I can call OpenGL 3D graphics hardware from Actionscript!
Love it.
Subscribe to:
Comments (Atom)