4 ms·
Image compression is complicated. Licencing topics, decompression cpu time, decompression memory use, decompression memory bandwidth use, compression time, is a
by JyrkiAlakuijala 7y ago
Image compression is complicated. Licencing topics, decompression cpu time, decompression memory use, decompression memory bandwidth use, compression time, is asymmetric slower but more efficient compression available, multi-threading, how much of the decoding can be shadowed during the data transfer, how many levels of progression are preferred, global image artefacts (such as banding) that don't appear in objective metrics, inter-frame copying ending up copying strange stuff around, red colors tend to not to be compressed with great success (there is no gamma compression for red in the eye, but in image formats there is), how much the image format is for the eyes and how much for the metrics, are the results the same on different platforms/implementations, is alpha supported, how good is the HDR modeling and its rather non-linear relations to gamma and colors in general, does the image format tend to preserve the quality of materials such as wood, marble, skin, cloth -- or replace them with cheap plastic imitation, some image formats work dramatically well at low BPP (<0.5) but start failing when compared to old JPEG at higher BPP (2.5+ BPP). Some image formats decode only a few megapixels per second which can be a disaster if your images happen to be in the 40 megapixel category.
Overall, it is a very complex landscape.
- jacobolus 7y ago> there is no gamma compression for red in the eye What does this mean? Do you have a citation to a source explaining whatever this is trying to say using precise standard color science terminology?
- airstrike 7y agoDon't you know every HN thread needs its standard armchair top comment rebuttal on how TFA is actually wrong and the author is naive and doesn't really understand the problem with all its intricacies?
- muizelaar 7y agoThe top comment rebuttal is written by the author of the WebP lossless, https://github.com/google/butteraugli https://github.com/google/butteraugli and https://github.com/google/pik https://github.com/google/pik so calling it armchairing doesn't really seem appropriate.
- coolspot 7y agoOuch
- ivalm 7y agoThis is the guy: https://ai.google/research/people/105344 https://ai.google/research/people/105344 https://scholar.google.com/citations?user=SnJNeR4AAAAJ&hl=en https://scholar.google.com/citations?user=SnJNeR4AAAAJ&hl=en Literally a world expert on the topic.
- ci5er 7y agoI know you are being a little tongue-in-cheek here, but I remember there was an Alvy-Ray Smith paper submitted about sampling scenes in time-and-space (and then filtering semi-super-sampled inputs), and it wasn't just the top comment. EVERYBODY knew better than ARS. Without having any DSP background. It was both amusing and sad.
- dang 7y agoPlease don't cross-examine people like this on HN. Threads here should be thoughtful conversations, not boxing matches. https://news.ycombinator.com/newsguidelines.html https://news.ycombinator.com/newsguidelines.html
- jacobolus 7y agoI am not “cross-examining” anyone. I am curious what this referred to, because I have never heard such a thing, and it doesn’t match my understanding of human color vision, a subject which I have studied fairly extensively. It was stated authoritatively, without any elaboration or evidence. I am asking for a citation so I can understand in more careful technical terms what claim is being made. (P.S. Having site moderators step in to discourage basic good-faith questions is a bad show.)
- dang 7y agoWell, moderation is guesswork and sometimes we guess wrong. Since you say your intent was to simply ask, I believe you. But it didn't read that way to me, which means it probably didn't to some others too. FWIW, the part that tipped me into thinking you were being aggressive was "whatever this is trying to say". That is the kind of thing people put into questions as a snarky swipe about someone being incoherent. Also, "do you have a citation to a source using precise standard science terminology" has a tinge of the same thing, since asking so bluntly and immediately for a citation is a way of not asking the commenter to explain what they meant themselves. Again, this is the sort of thing people do in internet comments when they're being haughty and dismissing the other person as ignorant. Especially a certain kind of technical haughtiness gets expressed this way. There's nothing wrong with asking for a citation, of course, but since it has become a sort of putdown trope, one's question should include enough information to convey intent. And in normal conversation one would typically ask the other person to explain what they were saying first.
- jacobolus 7y agoHow would you express the sentiment: “This statement seems incoherent based on what I know about the subject. To be charitable, this implies that it means something different than my interpretation, but was rendered confusing when simplified or expressed using terms pitched at laypeople. Could you please explain it using expert-targeted terminology or provide a citation to someone who does, so I can evaluate the claim on its merits instead of dismissing it as nonsense?” In a concise but friendly/productive way?
- TD-Linux 7y agoI'm pretty sure that's not literally true. For example, take a CRT, with its built in power law voltage-to-intensity response (the original gamma correction). If the eye didn't compress red, you'd see large shifts in the color of a white color ramp.
- BurningCycles 7y agoAny info on the future for PIK, is it on track to be assimilated into JPEG XL or will it remain a separate project ?