NHW Project - a free state-of-the-art image/video compression codec

[nhw_pulsar]

Hi john33,

Thank you very much for your answer.

Best,
Raphael

[excellentswordfight]

Originally Posted by nhw_pulsar

Hello,

I am slowly working on NHW, I had few ideas that on paper could/should work, but in practice turn out to not work at all... Hope I'll have other (better) ideas...

Else, very quickly, I have seen that the major trend for the MPEG new future video codec is, to summarize basically, to boost the latest MPEG DCT codecs with machine learning/neural networks, but I also had great conversations with 2 experts (of which 1 MPEG expert, I can not give the name) that also think that the next codec from MPEG should be low-complexity (in the era of 1080 and 4K) and so according to them MPEG could study NHW for their next codec as it has a competitive visual quality/compression at extremely lower complexity. They can not guarantee me that MPEG will consider/review NHW in an exploration phase (and even on an other side it will be quite difficult), but I thank them very much for their support.

Cheers,
Raphael

What market target are we talking about when we (they/MPEG) are discussing a low-complex codec? Visually lossless or highly compressed? For example JPEG-XS is not a DCT-codec, it can do visually lossless at up to 1:10 compression with only a few lines of delay, and do 1080p50 in realtime at a few watts (can be integrated in a sfp). And this what at least what I think of when it comes to low-complexity codecs, it should able to do real time encoding at high resolutions were a ASIC can do it running at a few watts. There are also mezz/DI use cases for low-complexity visually lossless codecs (e.g. the IMF(J2k)/prores-segment)

Or are you talking low complexity, high compression segment? E.g. mjpeg type of codec? So that would be the CCTV-segment.

[nhw_pulsar]

Originally Posted by excellentswordfight

What market target are we talking about when we (they/MPEG) are discussing a low-complex codec? Visually lossless or highly compressed? For example JPEG-XS is not a DCT-codec, it can do visually lossless at up to 1:10 compression with only a few lines of delay, and do 1080p50 in realtime at a few watts (can be integrated in a sfp). And this what at least what I think of when it comes to low-complexity codecs, it should able to do real time encoding at high resolutions were a ASIC can do it running at a few watts. There are also mezz/DI use cases for low-complexity visually lossless codecs (e.g. the IMF(J2k)/prores-segment)

Or are you talking low complexity, high compression segment? E.g. mjpeg type of codec? So that would be the CCTV-segment.

I would have understood that MPEG is looking for its new video codec at the same area as HEVC and VVC, but actually I really don't know, to be frank, I had very short conversation with the MPEG expert.

He told me that for him it is possible that MPEG considers NHW for its new video codec, as NHW has a competitive (high) compression at a very low complexity, but I absolutely don't know if many MPEG experts want low-complexity, or rather ever better PSNR/SSIM.

Cheers,
Raphael

[nhw_pulsar]

Hello,

Very quickly, for those interested, I have released the NHW v0.3.0-rc53 new version, always fine-tuning the nhw_kernel weights.

This new version seems to have a little more precision and so a better visual quality.

More at: https://nhwcodec.blogspot.com/

Cheers,
Raphael

[nhw_pulsar]

Hello,

Very quickly, for those interested, I have released the NHW v0.3.0-rc54 new version, always fine-tuning the nhw_kernel weights.

This new version seems to have a little more precision and so a better visual quality.

More at: https://nhwcodec.blogspot.com/

Cheers,
Raphael

[nhw_pulsar]

Hello,

Just to let you know that I start to be satisfied with the precision of the last version of NHW. Notably at -q14 high compression setting, I find now that NHW has a good and competitive precision, -while always keeping NHW efficient neatness-.

I don't know if I will be able to work on NHW in the next months, and concerning MPEG, I don't want to ask myself any more questions, because I understand nothing to it, 2 video compression experts (of which 1 MPEG member) told me that MPEG could consider NHW for their next video codec, and some other experts tell me that I am delirious, and I invented a pure lie because it is completely impossible that MPEG considers NHW... I would have expected a next press release/communication from MPEG where they could clarify if they could study/explore wavelets for their next video codec or not... This would help me to see clearer...

Cheers,
Raphael

[foxyshadis]

I applaud that after all these years you're still chasing your dream, but you really need to learn to sell your product, instead of just putting it out there and begging people to try it. And probably spend a lot more effort researching the industry, its standards, and the technical aspects of what gets adopted and what gets rejected. That's why many inventors hire agents who know how to work these groups; they don't come cheap, but if you aren't willing to put in the work to come to understand the needs and norms of companies and associations, you'll keep not getting anywhere. Most won't see a hobby project that doesn't even have a comprehensive demo page or spec as being worth a second glance. It's not the underlying technology -- that can be just about anything as long as it works -- it's the presentation.

FWIW, complete differences of opinions are very normal between MPEG members. They're people, after all. The discussions can get, ah, quite heated as engineers and stakeholders all hash out their differences.

[nhw_pulsar]

Hello,

Thank you very much for your answer and advice. Yes I am at the level-zero at selling my codec and that's a big problem, but for example making a good website/demo page and/or a clean spec with nice graphics, all this is out of my skills and especially out of my energy. Really the only I can do currently with my health is just to make some incremental small improvements of NHW...

Many thanks for confirming me that there are big disagreement between MPEG experts, I already noticed that because for their next video codec some experts very want to boost the latest MPEG codecs with neural networks, but for some others on the contrary it is essential now to be low-complexity...

Cheers,
Raphael

[lvqcl]

Originally Posted by nhw_pulsar

all this is out of my skills

Creating a good codec is also out of your skills though...

[nhw_pulsar]

Originally Posted by lvqcl

Creating a good codec is also out of your skills though...

Yes lossy compression is completely subjective/relative, with my "taste" I preferred more neatness, so I focus from the beginning to have a good neatness for NHW, and actually personally I am satisfied of the efficient neatness of the last version of NHW, always for my eyes, but I completely understand some other people are absolutely not sensitive to the neatness criteria and so for them NHW is awful and not interesting.

Cheers,
Raphael

[lvqcl]

No, I mean your codec is completely undocumented and doesn't support anything but 512x512 images (or has that changed?).

Also, a video codec without inter-frame compression? Really?

[nhw_pulsar]

Originally Posted by lvqcl

No, I mean your codec is completely undocumented and doesn't support anything but 512x512 images (or has that changed?).

Also, a video codec without inter-frame compression? Really?

Yes my codec is still undocumented and is still experimental for 512x512 resolution only, very sorry.

If NHW would be studied for the next MPEG video codec (which again is very not sure), yes it would be for the intra scheme, which would mean that MPEG would have to design an inter scheme that works with my intra wavelet technology (which makes things really more complex?).... If I understood clearly, MPEG could be mainly interested in exploring NHW (for the intra part) because it has a competitive high compression at a very low-complexity, and for some MPEG experts low-complexity has become a main essential criteria now... but again yes let's stay realistic the chances of NHW are very low.

Cheers,
Raphael

[rwill]

Originally Posted by nhw_pulsar

because it has a competitive high compression

Do you have a source for that?

[nhw_pulsar]

Originally Posted by rwill

Do you have a source for that?

Hello,

No sorry I don't have objective measurement/serious source for that, just I am testing NHW currently at -q11 and -q14 high compression settings, and I find that at these settings NHW results are visually more pleasant than AVIF and VVC for example, mainly because they have more neatness, but yes this is very subjective opinion...

And I wanted to clarify, this is not MPEG opinion neither, here is the very short conversation I had with a MPEG member (and I thank him very much for his time and answer), I asked him: "I have read that MPEG would like low-complexity for its next video codec, as NHW has a competitive high compression at very low-complexity, could MPEG study NHW for its next video codec?", and the MPEG member answered me: "Yes it is possible that we consider this."

But yes if the majority of MPEG members don't share this opinion then NHW won't be studied.

Cheers,
Raphael

[lvqcl]

Originally Posted by nhw_pulsar

and the MPEG member answered me: "Yes it is possible that we consider this."

Probably it was a very polite way to brush you off.

[nhw_pulsar]

Originally Posted by lvqcl

Probably it was a very polite way to brush you off.

Yes, seeing all the reactions I had, there is a possibility now that the MPEG member wanted to give me a kind answer and didn't want to depress me with the sad truth that NHW has no chance with MPEG, which many experts guarantee me...

Cheers,
Raphael

[Z2697]

What exactly is "neatness" that you've been talking about?

[nhw_pulsar]

Originally Posted by Z2697

What exactly is "neatness" that you've been talking about?

Hello,

I would define "neatness" in NHW as clean/neat edges/contours, for my eyes edges/contours are more blurred in AVIF and VVC for example, but yes in return they have a better fine details retention...

Cheers,
Raphael