3 ms·
You can't do that much on the decoding side (apart from the equivalent of passing the normally decoded result through a low percent img2img ML) But the encoder
by viraptor 3y ago
You can't do that much on the decoding side (apart from the equivalent of passing the normally decoded result through a low percent img2img ML)
But the encoders are already there: https://medium.com/@migel_95925/supercharging-jpeg-with-machine-learning-4471185d885d https://medium.com/@migel_95925/supercharging-jpeg-with-mach... https://compression.ai/ https://compression.ai/
- WithinReason 3y agoYou can more accurately invert the quantisation step
- viraptor 3y agoYou can't do it more accurately. You can make up expected details which aren't encoded in the file. But that's explicitly less accurate.
- adgjlsfhk1 3y agoIf the encoders know what model the decoders will be running, they can improve accuracy. You could pretty easily make a codec that doesn't encode high resolution detail if the decoder NN will interpolate it correctly.
- viraptor 3y agoThat's changing the encoder and sure, you could do that. But that's basically a new version of the format. It's not the JPEG we're using anymore + ML in decoder. It's JPEG-ML on both the encoder and decoder side. And with the speed that we adopt new image formats... That's going to take ages :(
- samus 3y agoThat makes sense if the goal is lossless compression. Since JPEG is lossy, it is sufficient to consider the Pareto front between quality, compressed size, and encoding/decoding performance.