It’s difficult to trace through the repo, but it looks like the compression ratios are computed by seeing how many pixel diffs were able to dropped. That’s inte

It’s difficult to trace through the repo, but it looks like the compression ratios are computed by seeing how many pixel diffs were able to dropped. That’s interesting, but the more interesting comparison would be to the average byte size of each frame in the compressed YouTube video. Without this, it’s hard to tell if the algorithm is an improvement relative to current methods. If the algorithm is lossy (small diffs are squeezed to zero) it probably should be compared to other lossy algorithms rather than lossless.