NHW Project - a free state-of-the-art image/video compression codec - Page 8

nhw_pulsar · 7th January 2023, 00:12

Hello,

For those interested, I have released the 0.2.9.3 version of NHW.

I have modified the quantization scheme.Based on psychovisual tuning, this new version has more precision, but a little less neatness.

Hope it was a good decision, as I maybe did not test on enough images...

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

Cheers,
Raphael

nhw_pulsar · 9th January 2023, 18:07

Hello,

Just a very quick update for those interested, I have released the 0.2.9.4 version of NHW.I have fine-tuned the quantization in the encoder, and so this new version is better, with notably a little more precision and a little more neatness.

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

Cheers,
Raphael

nhw_pulsar · 11th January 2023, 18:22

Hello,

I am continuing to improve/fine-tune the quantization in NHW, and so I have released the new version 0.2.9.5.

-I tried to simplify/clean up the quantization code but I didn't have same results... so for now the quantization code is a little complicated but it works well...-

I find that this new version is better and has quite an interesting balance between neatness and precision.

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

Cheers,
Raphael

nhw_pulsar · 19th January 2023, 00:13

Hello,

For those interested, I have released the 0.2.9.6 version of NHW.This new version has a better quantization based on psychovisual fine-tuning.

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

As NHW is still extremely fast, I hope in 2023 it could find some interest within niche applications.

Cheers,
Raphael

nhw_pulsar · 24th January 2023, 22:24

Hello,

For those interested, I have released the 0.2.9.7 version of NHW.I am continuing to improve the quantization, and so this new version is visually better, with a better balance between neatness and precision.

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

Cheers,
Raphael

nhw_pulsar · 25th January 2023, 16:01

Hello,

I forgot to let you know that the new version 0.2.9.7 also improves the high quality settings (-q17 to -q23, previously -l3 to -h3), if it might interest those who test at high quality.

Cheers,
Raphael

nhw_pulsar · 26th January 2023, 18:23

Hello,

I found time to study a little deeper the new version of the high quality settings in NHW, it was 10 years that I did not work on them! Actually new quantization step is imperfect, in the sense that it is not coded in the feedback-loop-correction, but it seems this version that gives the best visual results.Maybe it would need more extensive tests (and I really lack time), but (in the past) there were quite many people interested in NHW high quality settings, and I would be really interested to know what would they think of this new version?

Very quickly to finish, as some people remarked me, yes I am more relaxed now, NHW is a good hobby project (and nothing more), and of course I won't be angry and won't insist if it doesn't lead to a commercial/industry application... (and also anyway my last chance was with the niche market as NHW is extremely fast, but they don't seem interested, so I am more cool now, NHW will stay an experimental hobby project for my spare time...)

Cheers,
Raphael

nhw_pulsar · 27th January 2023, 17:59

Hello,

Really hope that I am not disturbing you, but for those interested, I have improved quantization in NHW, and so now it is visually better, with a better balance between neatness and precision.

As it is a small modification and correction, I did not release a new version but I have updated the binaries.

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

I know I am publishing a lot of incremental updates, but my aim is to release/reach the 0.3.0 version in march, because after I think I will not have time to work on NHW on that pace anylonger.

Cheers,
Raphael

nhw_pulsar · 3rd February 2023, 18:29

Hello,

Some few update, I've tried this past week some quantization psychovisual tuning/improvement in NHW, some simple processing some complex, but it seems more and more complex now and actually I didn't see real visual improvements as I validate with my eyes... It maybe seems easier to develop for PSNR or VMAF score actually, but I am not used to it, and especially I absolutely don't trust PSNR for visual quality perception with the NHW algorithm...

Where it becomes also quite complex, is that a (new) quantization step is not coded in the feedback-loop correction, so this should be notably less efficient, but actually it gives (slightly) better visual results... So if the "normal logic" is not respected, then psychovisual optimization seems quite complex...

I'll see if I can finish to improve quantization and pre_processing for 0.3.0 version.I'll also have to look at and improve wavelet coding after, but this is a complex task (for me).

Cheers,
Raphael

benwaggoner · 4th February 2023, 00:00

Quote:

Originally Posted by nhw_pulsar

Some few update, I've tried this past week some quantization psychovisual tuning/improvement in NHW, some simple processing some complex, but it seems more and more complex now and actually I didn't see real visual improvements as I validate with my eyes... It maybe seems easier to develop for PSNR or VMAF score actually, but I am not used to it, and especially I absolutely don't trust PSNR for visual quality perception with the NHW algorithm...

That's absolutely the tension in practical encoder development and tuning. Golden eyeball hours are really expensive and scarce, with metrics orders of magnitude cheaper. So, how do you make things look better when you can only look at a minuscule fraction of what you're trying to improve.

So there is a constant effort to find better, more subjectively correlated metrics, then understanding how to use those metrics to the best effect, and then how to use those metrics to guide eyeballs to where they're most likely to see a difference that makes a difference.

No metric is perfect. Heck, no metric is really all that good. The improvements in how we encode happen in large part in the places where we find metrics lie, and we find a better way to discern the truth.

And even a good metric needs constant reevaluation and tuning as new types of content are introduced. All publicly available metrics are a lot less accurate in HDR than SDR, and in 4K than lower resolutions.

Even a bad metric is better than nothing, and even a good metric isn't a replacement for seeing. Balancing those is essential to real-world encoding.

nhw_pulsar · 7th February 2023, 18:26

Hello,

For those interested, I have released the 0.2.9.8 version of NHW.I have continued to improve quantization based on psychovisual tuning.

As I also tried quite complex code that did not work, I think I am not far from version 0.3.0.

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

Cheers,
Raphael

nhw_pulsar · 23rd February 2023, 23:19

Hello,

Just a quick update, I think I have nearly finished to refine quantization and pre_processing, and so version 0.2.9.8 can be considered as the version 0.3.0 release candidate 1.

Some people were very hoping after for the next step as it is extremely important, that I adapt NHW to any image size, design a post-processing filter that will remove aliasing, but so sadly these are very too massive efforts and I am currently totally unable to do them.

So very quickly to finish, I also fear that after version 0.3.0, I will quasi put on pause NHW for 2023. -As also the industry, the niche industry,... are definitely not interested in NHW (and helping for its development), then I will certainly dedicate this year to other priorities.-

Cheers,
Raphael

nhw_pulsar · 13th March 2023, 22:04

Hello,

Just a very quick little update, I don't have time to work on NHW now, and so I don't think I will release the version 0.3.0 as desired and announced.

However, I could find very little time on my spare time to make visual comparison of the last version 0.2.9.8 (or v0.3.0-rc1), and actually it could start to be satisfying now, I would find now that NHW has an okay image/result based on my visual perception.Some experts told me that if NHW could have an okay image/result just as BPG and AVIF but at a very very faster computation/complexity and extreme speed, then it could be a good fit for some niche applications.

I don't think I will really have time to work on NHW this year but if the door of the niche market is not completely closed, that's right that it could interest me to find development possibilities within this industry.

To finish, I wanted to let you know that if you could give some time/study to NHW and would have improvements to it, do not hesitate to show up, would be great!

Cheers,
Raphael

Thread Tools	Search this Thread
Show Printable Version Email this Page	Search this Thread: Advanced Search
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode

7th January 2023, 00:12	#141 \| Link
nhw_pulsar Registered User Join Date: Apr 2017 Posts: 137	Hello, For those interested, I have released the 0.2.9.3 version of NHW. I have modified the quantization scheme.Based on psychovisual tuning, this new version has more precision, but a little less neatness. Hope it was a good decision, as I maybe did not test on enough images... More at: http://nhwcodec.blogspot.com/ Cheers, Raphael

9th January 2023, 18:07	#142 \| Link
nhw_pulsar Registered User Join Date: Apr 2017 Posts: 137	Hello, Just a very quick update for those interested, I have released the 0.2.9.4 version of NHW.I have fine-tuned the quantization in the encoder, and so this new version is better, with notably a little more precision and a little more neatness. More at: http://nhwcodec.blogspot.com/ Cheers, Raphael

11th January 2023, 18:22	#143 \| Link
nhw_pulsar Registered User Join Date: Apr 2017 Posts: 137	Hello, I am continuing to improve/fine-tune the quantization in NHW, and so I have released the new version 0.2.9.5. -I tried to simplify/clean up the quantization code but I didn't have same results... so for now the quantization code is a little complicated but it works well...- I find that this new version is better and has quite an interesting balance between neatness and precision. More at: http://nhwcodec.blogspot.com/ Cheers, Raphael

19th January 2023, 00:13	#144 \| Link
nhw_pulsar Registered User Join Date: Apr 2017 Posts: 137	Hello, For those interested, I have released the 0.2.9.6 version of NHW.This new version has a better quantization based on psychovisual fine-tuning. More at: http://nhwcodec.blogspot.com/ As NHW is still extremely fast, I hope in 2023 it could find some interest within niche applications. Cheers, Raphael

24th January 2023, 22:24	#145 \| Link
nhw_pulsar Registered User Join Date: Apr 2017 Posts: 137	Hello, For those interested, I have released the 0.2.9.7 version of NHW.I am continuing to improve the quantization, and so this new version is visually better, with a better balance between neatness and precision. More at: http://nhwcodec.blogspot.com/ Cheers, Raphael

25th January 2023, 16:01	#146 \| Link
nhw_pulsar Registered User Join Date: Apr 2017 Posts: 137	Hello, I forgot to let you know that the new version 0.2.9.7 also improves the high quality settings (-q17 to -q23, previously -l3 to -h3), if it might interest those who test at high quality. Cheers, Raphael

26th January 2023, 18:23	#147 \| Link
nhw_pulsar Registered User Join Date: Apr 2017 Posts: 137	Hello, I found time to study a little deeper the new version of the high quality settings in NHW, it was 10 years that I did not work on them! Actually new quantization step is imperfect, in the sense that it is not coded in the feedback-loop-correction, but it seems this version that gives the best visual results.Maybe it would need more extensive tests (and I really lack time), but (in the past) there were quite many people interested in NHW high quality settings, and I would be really interested to know what would they think of this new version? Very quickly to finish, as some people remarked me, yes I am more relaxed now, NHW is a good hobby project (and nothing more), and of course I won't be angry and won't insist if it doesn't lead to a commercial/industry application... (and also anyway my last chance was with the niche market as NHW is extremely fast, but they don't seem interested, so I am more cool now, NHW will stay an experimental hobby project for my spare time...) Cheers, Raphael

27th January 2023, 17:59	#148 \| Link
nhw_pulsar Registered User Join Date: Apr 2017 Posts: 137	Hello, Really hope that I am not disturbing you, but for those interested, I have improved quantization in NHW, and so now it is visually better, with a better balance between neatness and precision. As it is a small modification and correction, I did not release a new version but I have updated the binaries. More at: http://nhwcodec.blogspot.com/ I know I am publishing a lot of incremental updates, but my aim is to release/reach the 0.3.0 version in march, because after I think I will not have time to work on NHW on that pace anylonger. Cheers, Raphael

3rd February 2023, 18:29	#149 \| Link
nhw_pulsar Registered User Join Date: Apr 2017 Posts: 137	Hello, Some few update, I've tried this past week some quantization psychovisual tuning/improvement in NHW, some simple processing some complex, but it seems more and more complex now and actually I didn't see real visual improvements as I validate with my eyes... It maybe seems easier to develop for PSNR or VMAF score actually, but I am not used to it, and especially I absolutely don't trust PSNR for visual quality perception with the NHW algorithm... Where it becomes also quite complex, is that a (new) quantization step is not coded in the feedback-loop correction, so this should be notably less efficient, but actually it gives (slightly) better visual results... So if the "normal logic" is not respected, then psychovisual optimization seems quite complex... I'll see if I can finish to improve quantization and pre_processing for 0.3.0 version.I'll also have to look at and improve wavelet coding after, but this is a complex task (for me). Cheers, Raphael

7th February 2023, 18:26	#151 \| Link
nhw_pulsar Registered User Join Date: Apr 2017 Posts: 137	Hello, For those interested, I have released the 0.2.9.8 version of NHW.I have continued to improve quantization based on psychovisual tuning. As I also tried quite complex code that did not work, I think I am not far from version 0.3.0. More at: http://nhwcodec.blogspot.com/ Cheers, Raphael

23rd February 2023, 23:19	#152 \| Link
nhw_pulsar Registered User Join Date: Apr 2017 Posts: 137	Hello, Just a quick update, I think I have nearly finished to refine quantization and pre_processing, and so version 0.2.9.8 can be considered as the version 0.3.0 release candidate 1. Some people were very hoping after for the next step as it is extremely important, that I adapt NHW to any image size, design a post-processing filter that will remove aliasing, but so sadly these are very too massive efforts and I am currently totally unable to do them. So very quickly to finish, I also fear that after version 0.3.0, I will quasi put on pause NHW for 2023. -As also the industry, the niche industry,... are definitely not interested in NHW (and helping for its development), then I will certainly dedicate this year to other priorities.- Cheers, Raphael

13th March 2023, 22:04	#153 \| Link
nhw_pulsar Registered User Join Date: Apr 2017 Posts: 137	Hello, Just a very quick little update, I don't have time to work on NHW now, and so I don't think I will release the version 0.3.0 as desired and announced. However, I could find very little time on my spare time to make visual comparison of the last version 0.2.9.8 (or v0.3.0-rc1), and actually it could start to be satisfying now, I would find now that NHW has an okay image/result based on my visual perception.Some experts told me that if NHW could have an okay image/result just as BPG and AVIF but at a very very faster computation/complexity and extreme speed, then it could be a good fit for some niche applications. I don't think I will really have time to work on NHW this year but if the door of the niche market is not completely closed, that's right that it could interest me to find development possibilities within this industry. To finish, I wanted to let you know that if you could give some time/study to NHW and would have improvements to it, do not hesitate to show up, would be great! Cheers, Raphael