3 ms·
This is really cool, but in other comments I'm reading that good JPEG encoders and decoders depend on knowing the other. That you can design a better JPEG enco
by crazygringo 6y ago
This is really cool, but in other comments I'm reading that good JPEG encoders and decoders depend on knowing the other.
That you can design a better JPEG encoder for a known reference decoder (which decoded version best matches the original), and you can likewise design a better JPEG decoder for a known reference encoder.
But now that there are both encoders and decoders trying to be more efficient and non-blocky respectively... it feels like a mess. Even more so because a decoder never knows which encoder was used, and an encoder never knows which decoder.
Is there any solution to this mess? Is there a best practice here? Since most JPEG's are probably viewed on the web, do all browsers share the same decoder, and so effort is best spent on the encoder? Or even while encoder improvements are made specifically for the algorithms browsers use, can improvements to decoders still be made that remain improvements, instead of e.g. "overfitting" and actually making things worse?
This just feels like a bit of an "improvements race" on both sides which leaves me a little dizzy in trying to understand it...
- IgorPartola 6y agoThe obvious improvement is to use a better format than JPEG. Maybe something that can include information about the encoding process so the decoder can use that instead of guessing. Or maybe even a non-lossy format so we can focus more on data compression and less on visual compression.