Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

H.266/Versatile Video Coding (VVC)

newsletter.fraunhofer.de

Post reply on HN
Print view View original

431–435 of 435 posts

Re: H.266/Versatile Video Coding (VVC)

#431

Post by robomartin »

somethingsome wrote: post #419
robomartin wrote: post #175

Among other things, I have worked with and developed technology in the uncompressed professional imaging domain for decades. One of the things I always watch out for is precisely the terminology and language used in this release: "for equal perceptual quality" Put a different way: We can fool your eyes/brain into thinking you are looking at the same images. For most consumer use cases where the objective is to view i…

Don't you think that the iterated convolution process in neural networks is, in a measure, able to overlook this kind of 'visual trickery' ? I can imagine that the network is not able to perform well if you change the color profile of the input when you trained on another one, but small texture attenuations, diminished chroma components, etc. may not be as important when the image is downsampled and split a huge numb…

Here's one way to look at it: Contrast and well defined edges can be important in feature extraction. Our vision system, on the other hand, can do just fine with less information in the high frequencies (where edges live).

Re: H.266/Versatile Video Coding (VVC)

#433

Post by somethingsome »

robomartin wrote: post #431
somethingsome wrote: post #419

Earlier quoted context omitted.

Don't you think that the iterated convolution process in neural networks is, in a measure, able to overlook this kind of 'visual trickery' ? I can imagine that the network is not able to perform well if you change the color profile of the input when you trained on another one, but small texture attenuations, diminished chroma components, etc. may not be as important when the image is downsampled and split a huge numb…

Here's one way to look at it: Contrast and well defined edges can be important in feature extraction. Our vision system, on the other hand, can do just fine with less information in the high frequencies (where edges live).

I see your point, and I'm not particularly defending neural networks, but IMO, nothing prevents a network to generate a kernel able to detect 'fuzzy edges' and to refine it to an edge after some convolutions. So, if the input images are always consistent between them and with the input images for inference, I think the problem may be diminished (?), even if, as you say, we introduce some misclassification error. Obviously, to have the guarantee that all the input images are generated in the same way is a very strong condition difficult to achieve.

Re: H.266/Versatile Video Coding (VVC)

#434

Post by robomartin »

somethingsome wrote: post #433
robomartin wrote: post #431

Earlier quoted context omitted.

Here's one way to look at it: Contrast and well defined edges can be important in feature extraction. Our vision system, on the other hand, can do just fine with less information in the high frequencies (where edges live).

I see your point, and I'm not particularly defending neural networks, but IMO, nothing prevents a network to generate a kernel able to detect 'fuzzy edges' and to refine it to an edge after some convolutions. So, if the input images are always consistent between them and with the input images for inference, I think the problem may be diminished (?), even if, as you say, we introduce some misclassification error. Obvi…

From my perspective, the only way to get there is if AI practitioners make a paradigm shift towards encoding understanding rather than making classifier systems trained with massive data sets. The classification approach has a very real asymptotic limit on what can be achieved. You can train NN's using large data sets on some domains but not all domains. Just think about what a dog can do, even just a puppy. We are nowhere near to that. Not even close. This is because our AI classifies without understanding.

I have books on AI that are thirty years old. I think I can say they cover somewhere between 80% and 90% (if not more) of what AI is today. The difference is computing that is thousands, millions, of times faster, massive amounts of storage, etc. In other words, one could very well argue we haven't done much in 30 years other than build faster computers.

Re: H.266/Versatile Video Coding (VVC)

#435

Post by somethingsome »

robomartin wrote: post #434
somethingsome wrote: post #433

Earlier quoted context omitted.

I see your point, and I'm not particularly defending neural networks, but IMO, nothing prevents a network to generate a kernel able to detect 'fuzzy edges' and to refine it to an edge after some convolutions. So, if the input images are always consistent between them and with the input images for inference, I think the problem may be diminished (?), even if, as you say, we introduce some misclassification error. Obvi…

From my perspective, the only way to get there is if AI practitioners make a paradigm shift towards encoding understanding rather than making classifier systems trained with massive data sets. The classification approach has a very real asymptotic limit on what can be achieved. You can train NN's using large data sets on some domains but not all domains. Just think about what a dog can do, even just a puppy. We are n…

I don't think that neural networks are the right framework to achieve general purpose artificial intelligence (AGI). And indeed, the AI field may need a paradigm shift to achieve higher classification goals. I believe that probabilistic neural networks may be an interesting extension toward general purpose AI, even though this kind of networks need even more data.

If we take the example of a puppy, it seems to generalize pretty well using something like one-shot learning, but is it? I cannot confirm for sure how much data a puppy has already digested before being able to do what we could call "one shot learning". So maybe, the exposure to data is already there, waiting for a specialization toward a particular task.

Giving the ability to a network to be probabilistic enable it to do inference using uncertainty, which is clearly a neat feature when you are gravitating toward AGI for scene understanding.

In the case of video compression, scene understanding may introduce more artifacts IMO: Even if the scene is captured with high end cameras, on a pixel level basis, the edges will never be perfectly neat. I think this will decrease the ability of any network to "understand" which object is at the edges, this results in low classification rates on them, resulting in bad compression/decompression quality (?) for features that are important to the human eye.

All in all, I'm not sure that NN are the right tool for this kind of problems. But we are diverging from the main subject VVC, Thanks for the very interesting comments :)

Post reply on HN