Preparation for Banff

Posted: January 6, 2016 at 7:54 pm


I’ve been doing a little more work trying to get things ready for my residency at the Banff Centre starting next week. In attempting to reduce the amount of noise (in the form of the presence of large percepts popping in and out for each frame), I changed the code that generates percepts to ignore regions whose area is over a particular threshold. The result is exactly the opposite of what I expected where all the visible percepts are quite large (see below).

I double checked and indeed the code is correct in only keeping percepts with an area below a threshold. I have since changed the code that tells the renderer what percepts should be presented and where so that it will scale percepts according to the ratio between the area of the cluster centre and the region being represented. Unfortunately this will take 85 hours, and thus I will not have gotten as far as I had hoped before getting to Banff.

It seems the real problem is that there is too much diversity of regions, and thus they end up holding highly different content. Either I can increase the number of segments in each frame, or I can filter large percepts before clustering. Since the film already has 30million regions, making even more is unwise considering clustering already took 30 days (aside, I realized I did not have TBB installed, and thus kmeans in opencv could now be 4 times faster on this quad machine). I did a test with a different threshold and ended up with 6 times more regions. It seems filtering regions before kmeans is the best idea, but since my residency is only 5 weeks, I’ll have to work with a subset of the film so I can do the clustering in under one week. Some of these images are still quite interesting, but far too abstract considering these are perceptual reconstructions meant to reproduce images from the film:

store-0000005 store-0008847 store-0009912 store-0012675 store-0013898 store-0016402 store-0025214 store-0039073