Thanks, yes it would be easy. You should modify the file https://github.com/matteo-ronchetti/Pointillism/blob/master/main.py as follows:
reduce the batch_size on line 58
At the end of the for loop of line 61 save the image (stored in the variable res)
Then you can combine all the saved images to make a video
In my personal experience I spend many days setting up an experiment and doin some preliminary small scale experiments. For this kind of experimentation a normal desktop computer with a decent GPU is ok. Then when I need to do serious experiments I use AWS spot instances (3x cheaper) with the advantage that I can run many experiments in parallel.
Nice experiment! Let me share 2 taughts on it:
I think that for estimating the covariance matrix 20 samples (a batch size of 20) is not enough. You will end up with a noisy estimate of the mean/covariance. This noisy estimate will make you apply different linear transformations to different points in space and I don’t know…