Welcome to Doom9's Forum, THE in-place to be for everyone interested in DVD conversion.

Before you start posting please read the forum rules. By posting to this forum you agree to abide by the rules.

 

Go Back   Doom9's Forum > Video Encoding > New and alternative video codecs

Reply
 
Thread Tools Search this Thread Display Modes
Old 11th August 2021, 21:45   #81  |  Link
nhw_pulsar
Registered User
 
Join Date: Apr 2017
Posts: 171
Quote:
Originally Posted by benwaggoner View Post
I believe that project is focused on lowering ENCODING time for video codecs. This would be stuff like the ASICs YouTube is building. Pixels-per-second/watt is quite important for UGC like FaceBook and YouTube, as they do huge volumes of encoding of files, of which most get watched <5 times, and being free customer quality expectations are much lower. It's a very different market than for premium content.

Unless you can have a compelling argument for why decode would be at least as low power as AV1 software decoders, and a realistic plan for encoding much faster than AV1 at similar quality, it's probably not a good fit. At a minimum you'd need to show how interframe prediction would work, and work efficiently.
Ok, thanks for the clarification.I did not understand it was energy saving of current codecs (with better hardware), I thought it was new algorithms... But yes, I am not expert of inter-frame coding, so I can not suggest an efficient scheme in conjunction with NHW wavelet intra.I however think that with some investment from the industry if they had the will, this could be achievable...

Cheers,
Raphael
nhw_pulsar is offline   Reply With Quote
Old 12th August 2021, 23:28   #82  |  Link
benwaggoner
Moderator
 
Join Date: Jan 2006
Location: Portland, OR
Posts: 4,752
The challenge is that wavelet transforms for interframe encoding have been tried many times, and have never resulted in anything competitive with block-based encoding and motion estimation. So unless you have a promising novel ideal for how to do it better this time, it's unlikely you'll get much interest.

I think Monty from Xiph did some posts about this ages ago.
__________________
Ben Waggoner
Principal Video Specialist, Amazon Prime Video

My Compression Book
benwaggoner is offline   Reply With Quote
Old 15th August 2021, 21:33   #83  |  Link
nhw_pulsar
Registered User
 
Join Date: Apr 2017
Posts: 171
Quote:
Originally Posted by benwaggoner View Post
The challenge is that wavelet transforms for interframe encoding have been tried many times, and have never resulted in anything competitive with block-based encoding and motion estimation. So unless you have a promising novel ideal for how to do it better this time, it's unlikely you'll get much interest.

I think Monty from Xiph did some posts about this ages ago.
Actually, for example overlapped block motion estimation/compensation OBMC/OBME with wavelets have been reported in the 2004-2008 period to be as good or subjectively better than H.264/AVC, but since then not much work has been made on them to the big contrary of ME/MC used in HEVC/VVC.

Actually, I would really like to work on inter-frame coding for a wavelet video codec, I have also some ideas to test, but for that I must first adapt for example NHW to any image size (I will not develop for 512x512 video sequences), all that is a lot of work, very too much for my free time, and so if a company/organization is willing to explore the NHW new technology and to sponsorize/back me, it would be great!

Cheers,
Raphael
nhw_pulsar is offline   Reply With Quote
Old 4th September 2021, 19:33   #84  |  Link
nhw_pulsar
Registered User
 
Join Date: Apr 2017
Posts: 171
Hello,

I have finally released the 0.2.0 version of NHW.This one has a better neatness and a better precision.

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

As usual, the new entropy coding schemes have not been totally optimized, and so we can save at least 2.5KB (and more) per .nhw compressed file.

Any feedback welcome!

Cheers,
Raphael
nhw_pulsar is offline   Reply With Quote
Old 6th October 2021, 18:50   #85  |  Link
nhw_pulsar
Registered User
 
Join Date: Apr 2017
Posts: 171
Hello,

I have (slowly) worked these last couple weeks on the psychovisual model of NHW because some people reproach me that it washes details out... so I restored a little more details but this consumes some bits and actually it turned out that it also decreases neatness, and so when you compare at the same compression, actually I find the current version visually better because it has a better neatness despite washing out details.

So I would like to put forward that the main concept of NHW is neatness (and also ultra-fast encoding/decoding), it's a total different approach from AV1, VVC,... that relies on precision, but I start to find (and this is my personal opinion based on my visual appreciation) that the current version of NHW could prove the viability of neatness and of the NHW technology.We can still improve it, but on the other side I am more and more pessimistic that the industry would like to consider NHW, as for example I have totally lost contacts, with AOM, MPEG, some major companies,... since 2 years now.

Again, I write this message because I believe in my technology (and I am certainly the only one currently) and so I would like to develop it, and notably change my very difficult current structure with absolutely no support that only allows me to work on NHW on my very few spare time.If you think you can help me, like for example renewing contact with the industry that is very cold with the NHW technology, please do not hesitate to answer, it would be great!

Cheers,
Raphael
nhw_pulsar is offline   Reply With Quote
Old 2nd December 2021, 20:41   #86  |  Link
nhw_pulsar
Registered User
 
Join Date: Apr 2017
Posts: 171
Hello,

I know there is no community around NHW, but I would have liked to know what you would like to see improved in the NHW codec.I know its results are not perfect for now compared to high-end AV1/VVC codecs for example, but could be ok giving its extremely fast speed.-However on high quality images, I notably like its good results because of its neatness.-

Regarding things to improve, there is the problem of aliasing, I have started some quick experiments as I mainly find that aliasing appears in NHW along "low-contrast edges", but I have also seen some very impressive results of directional filters like the ones used in CDEF filter in AV1, because they estimate the "strong edge direction" and align all the pixels on that "strong edge direction" which could depixelize/remove aliasing.I have also started to implement a banding detector (with its main analysis on the 128x128 wavelet DC image) in order to remove banding, and it's only the beginning but I have promising results.I have also started to improve the entropy coding scheme of the wavelet DC part, for now I have a very fast processing that gives another 1KB saving on that part in average, but I have an error on some few images, I don't think I contradict the compression and information theory principles, I think it is more a corner case that I did not see for now and so that I don't handle, but it takes a lot of time... I'll also have to better handle the chroma channels and notably add chroma from luma technique.

Some people also tell me that the priority now is to adapt NHW to any image size, but actually this is the boring task for me currently, and I am a little reluctant for now.

All that to let you know that I really like more to work on my codec, and so I would like now to work full-time on it again, but there is always the big problem of the salary at the end of the month.As the industry doesn't show interest in NHW, I would like to ask sponsoring for this project on the Internet.But some people told me that I failed to create a community around NHW, and with just 27 followers on GitHub and 8 followers on Twitter (that's very weak), I can not hope more than 10dollars/per month of sponsoring/donation.

Do you think this solution/sponsoring is possible? Or on the contrary, NHW will be just a spare time hobby, because there is nothing more to expect out of it?

Any help/opinion is very welcome!

Cheers,
Raphael

Last edited by nhw_pulsar; 2nd December 2021 at 20:44.
nhw_pulsar is offline   Reply With Quote
Old 2nd December 2021, 21:14   #87  |  Link
Selur
Registered User
 
Selur's Avatar
 
Join Date: Oct 2001
Location: Germany
Posts: 7,259
No clue whether this will find sponsors, but it for most folks will not find it even worth looking at since they don't work with 512x512 resolutions.
That seems more like a resolution you might use for icons or textures rather than video or images nowadays.
__________________
Hybrid here in the forum, homepage
Selur is offline   Reply With Quote
Old 2nd December 2021, 21:23   #88  |  Link
nhw_pulsar
Registered User
 
Join Date: Apr 2017
Posts: 171
Quote:
Originally Posted by Selur View Post
No clue whether this will find sponsors, but it for most folks will not find it even worth looking at since they don't work with 512x512 resolutions.
That seems more like a resolution you might use for icons or textures rather than video or images nowadays.
Yes, NHW started as a 512x512 proof-of-concept on my spare time.But actually now to turn it into any image size, it will require me full-time work.And so I hoped that some people could be interested in the results of that 512x512 demo version/proof-of-concept, and could help me to develop it into any image size.I was just thinking of sponsoring on the Internet to be able to do that full-time job and also as the industry is not interested.But it seems complicated...

Cheers,
Raphael
nhw_pulsar is offline   Reply With Quote
Old 4th December 2021, 11:26   #89  |  Link
rwill
Registered User
 
Join Date: Dec 2013
Posts: 343
Quote:
Originally Posted by nhw_pulsar View Post
Any help/opinion is very welcome!
I have an opinion.

Lossy image wavelet codecs are a dime a dozen. In certain CS courses students even have to make their own separate implementation to get accepted to the course exam, thats how common they are. I looked at your code. There is nothing special or innovative. Its just one more lossy format with serious restrictions. Over two decades ago when I was doing my degree thing I implemented a couple of these too, heck, even with arbitrary resolution and some were video codecs with complex inter coding. I did this for fun and out of interest and then moved on to other things. I looked at your code. Lots of things are hard coded and limiting when they could be done dynamically. Your neatness metric is just some distortion measure based on the transform used it seems. My estimate is that the current NHW implementation can probably be outclassed by gluing stock code for a db 9/7 transform, some subband quantization and some predictive arithmetic entropy coding together.

Implementing some wavelet or dct based image codec - for fun - ok.
But please - drop the idea that someone else could have a use for it or even that you can make money with it.

Sorry.
rwill is offline   Reply With Quote
Old 4th December 2021, 12:18   #90  |  Link
nhw_pulsar
Registered User
 
Join Date: Apr 2017
Posts: 171
Quote:
Originally Posted by rwill View Post
I have an opinion.

Lossy image wavelet codecs are a dime a dozen. In certain CS courses students even have to make their own separate implementation to get accepted to the course exam, thats how common they are. I looked at your code. There is nothing special or innovative. Its just one more lossy format with serious restrictions. Over two decades ago when I was doing my degree thing I implemented a couple of these too, heck, even with arbitrary resolution and some were video codecs with complex inter coding. I did this for fun and out of interest and then moved on to other things. I looked at your code. Lots of things are hard coded and limiting when they could be done dynamically. Your neatness metric is just some distortion measure based on the transform used it seems. My estimate is that the current NHW implementation can probably be outclassed by gluing stock code for a db 9/7 transform, some subband quantization and some predictive arithmetic entropy coding together.

Implementing some wavelet or dct based image codec - for fun - ok.
But please - drop the idea that someone else could have a use for it or even that you can make money with it.

Sorry.
Ok, thank you for your professional opinion.It's a very precious information for me, because the industry for example didn't dare to be that direct and that frank, and so there was a doubt for me, but now it's very clear, NHW will just stay a spare time hobby.

Do you bother that I sometimes post my updates on that professional forum each 6 months but clearly I will stop speaking about sponsoring?

Cheers,
Raphael
nhw_pulsar is offline   Reply With Quote
Old 4th December 2021, 17:55   #91  |  Link
rwill
Registered User
 
Join Date: Dec 2013
Posts: 343
Hey go ahead, you can even write about sponsoring if you like. Maybe I missed something in NHW and there is some novel idea somewhere inside.

To view things from a slightly different angle.....

What you are doing is research. A bread-less art if done without institutional support. Most wavelet compression research is done in academia at universities or research institutes. Some notable wavelet image compression formats that 'made it' in the last years are Jpeg2K, Dirac and Jpeg-XS but these pale in comparison to the formats the ITU put out. It is hard to compete against these hundreds of researchers alone so to make a living off it you would have to join them somehow.

I would try to determine what the state of the art of wavelet based image compression is, try to sort my stuff into that and write a paper when I think I have a clear breakthrough. Problem is, publishing a paper is hard and expensive without institutional support too.

This forum ? Most people on this forum have about the same knowledge about compression technology that a grandma has about modern Web Technologies when she uses them to post on Facebook with the edge browser. I would not expect technical feedback when you post about your wavelet research here. Academia would be a better place.
rwill is offline   Reply With Quote
Old 4th December 2021, 19:19   #92  |  Link
nhw_pulsar
Registered User
 
Join Date: Apr 2017
Posts: 171
Quote:
Originally Posted by rwill View Post
Hey go ahead, you can even write about sponsoring if you like. Maybe I missed something in NHW and there is some novel idea somewhere inside.

To view things from a slightly different angle.....

What you are doing is research. A bread-less art if done without institutional support. Most wavelet compression research is done in academia at universities or research institutes. Some notable wavelet image compression formats that 'made it' in the last years are Jpeg2K, Dirac and Jpeg-XS but these pale in comparison to the formats the ITU put out. It is hard to compete against these hundreds of researchers alone so to make a living off it you would have to join them somehow.

I would try to determine what the state of the art of wavelet based image compression is, try to sort my stuff into that and write a paper when I think I have a clear breakthrough. Problem is, publishing a paper is hard and expensive without institutional support too.

This forum ? Most people on this forum have about the same knowledge about compression technology that a grandma has about modern Web Technologies when she uses them to post on Facebook with the edge browser. I would not expect technical feedback when you post about your wavelet research here. Academia would be a better place.
Ok, yes I humbly think that I prefer wavelet compression research than software development: adapt NHW to any arbitrary image size...

Actually I don't really know the current state-of-the-art wavelet compression research.I however read some impressive academic papers, but I don't have access to the corresponding source code or binaries.The latest state-of-the-art in wavelet compression I have is Rududu Image Codec, which represented the latest evolution of SPIHT-like, zerotree-like wavelet coding, but it is from 2008!...-NHW doesn't use that coding scheme for the info.-

It would be so great to be in contact/work for standard bodies (AOM,MPEG,JPEG,...) or academia, but to be honest my attempts at it were very discouraging... so I have a little lost hope now... but anyway if you would have any advice, do not hesitate to let me know.

Cheers,
Raphael
nhw_pulsar is offline   Reply With Quote
Old 7th December 2021, 17:52   #93  |  Link
nhw_pulsar
Registered User
 
Join Date: Apr 2017
Posts: 171
Hi @rwill,

Actually that's right that I don't know the current state of wavelet compression research that happens in academia as you indicated me.I have read very impressive results from 2005 (same PSNR as AVIF/VVC today!) of wavelet compression projects based on enhanced zerotree coding (and enhanced predictive & adaptive arithmetic coding), one of these technologies was called CEZW if I remember correctly.That's right that the academia is maybe focusing on these (zerotree-like) technologies turned toward PSNR and SSIM goals, and so maybe NHW will not represent an interest for them.-NHW is however extremely faster in return...-

So to know if NHW could have an interest for academia, I would like to contact them and try to have a further discussion about the NHW technology.As I don't have contacts within the academia, could you suggest me rwill someone that I could contact there?

Many thanks.
Cheers,
Raphael
nhw_pulsar is offline   Reply With Quote
Old 7th December 2021, 20:09   #94  |  Link
lvqcl
Registered User
 
Join Date: Aug 2015
Posts: 293
Do you have anything to show them?
lvqcl is offline   Reply With Quote
Old 7th December 2021, 20:25   #95  |  Link
nhw_pulsar
Registered User
 
Join Date: Apr 2017
Posts: 171
Quote:
Originally Posted by lvqcl View Post
Do you have anything to show them?
That's the point!

NHW has a new fast wavelet transform, a new multistage residual coding, a new "psychovisual model", 3 new compression/entropy coding schemes, and some other things.

But will it be interesting for the academia? Is it worth to disturb them? That's the point!

Cheers,
Raphael
nhw_pulsar is offline   Reply With Quote
Old 7th December 2021, 20:35   #96  |  Link
lvqcl
Registered User
 
Join Date: Aug 2015
Posts: 293
Quote:
Originally Posted by nhw_pulsar View Post
NHW has a new fast wavelet transform, a new multistage residual coding, a new "psychovisual model", 3 new compression/entropy coding schemes, and some other things.
Where can I read about them?
lvqcl is offline   Reply With Quote
Old 7th December 2021, 20:54   #97  |  Link
nhw_pulsar
Registered User
 
Join Date: Apr 2017
Posts: 171
Quote:
Originally Posted by lvqcl View Post
Where can I read about them?
Actually, I didn't write paper about them.They are just in the source code of the NHW GitHub.

It would be so great to have some kind of academia support to write and publish paper, if it is worthwhile.
nhw_pulsar is offline   Reply With Quote
Old 7th December 2021, 23:32   #98  |  Link
nhw_pulsar
Registered User
 
Join Date: Apr 2017
Posts: 171
Quote:
Originally Posted by lvqcl View Post
Where can I read about them?
Sorry if I misunderstood your request.Yes there is no paper or documentation of NHW processings/technology.

But however in the case if you would really like some deeper details on some NHW processings, I can give you some explanation on your email as first solution.

Cheers,
Raphael
nhw_pulsar is offline   Reply With Quote
Old 8th December 2021, 09:11   #99  |  Link
jpsdr
Registered User
 
Join Date: Oct 2002
Location: France
Posts: 2,309
Maybe it could be a good time to write some kind of documentation. It will be the first thing people who may be interested will ask for. If you want to try to promote your work, you'll need to have some documentation which explain your work, and where or why it's innovative.
__________________
My github.
jpsdr is offline   Reply With Quote
Old 8th December 2021, 10:13   #100  |  Link
nhw_pulsar
Registered User
 
Join Date: Apr 2017
Posts: 171
Quote:
Originally Posted by jpsdr View Post
Maybe it could be a good time to write some kind of documentation. It will be the first thing people who may be interested will ask for. If you want to try to promote your work, you'll need to have some documentation which explain your work, and where or why it's innovative.
Yes, that would be better.

But actually, for example I don't know the software used by the academia to make publications, especially the one used to make good/clean graphics... this last task can take some time for me, but I'll have to find a solution as you rightly suggested.

Cheers,
Raphael
nhw_pulsar is offline   Reply With Quote
Reply

Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT +1. The time now is 08:06.


Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2024, vBulletin Solutions Inc.