Machine Learning / AI and the Future of Compression

[jcs]

While fractal and wavelet compression made a splash a few years ago (and wavelet is still used today in JPEG2000 and Redcode RAW), there's a lot more to be discovered based on generative mathematics and machine learning.

Here's one cool example: http://www.wave.one/icml2017/

 

[ntblowz]

interesting link, will it make video file size even much smaller?

[jcs]

Yeah, generative compression applied to moving images (video) will result in a massive reduction in file sizes vs. the current macroblock DCT methods.

[ND64]

But 0.08 bit per pixel is around 20Mb/s for 4k@30. I'm sure h.265 is already producing a pretty clear image at this data rate. 

[jcs]

2 hours ago, Eric Calabros said:

But 0.08 bit per pixel is around 20Mb/s for 4k@30. I'm sure h.265 is already producing a pretty clear image at this data rate. 

That would be using all I-frames (no compressed motion deltas). To really leapfrog H.265, they've got to apply the generative encoding concept to inter-frame compression. A hybrid concept would be to use generative encoding for I-frames, and use concepts from H.265 for inter-frame encoding.