Want to see wavelet divx

[sumpm1]

I have seen a really crappy codec called wavc that uses wavelet compression. The codec didnt work very well, and it was very slow to encode, but the picture quality was good with very very good compression(small files) I have seen this toppic discussed before, but I wonder if this could be implemented into divx?

[-h]

It could be, yes, because wavelets are supported in the MPEG4 spec (to varying degrees). However, no one really has any intention of working on them yet.

Unless you're volunteering :)

-h

[Nic]

I personally have a hate of people bringing up wavelets who don't understand them & asking for them to be implemented. I haven't seen a good wavelet implementation yet & although they are very promising in theory (they're are even faster than DCT in theory) they are very complex, so please...no more questions here (unless you've at least read everything on the tarkin mailist, looked through a few university websites & have a basic knowledge of the MPEG-4 profiles)

No Offense though :)

Cheers,
-Nic

[Koepi]

Hm. My experience is that the wavelet codecs produce output that looks like shit....

Just my 2 € cent.

Regards,
Koepi

[sumpm1]

Sorry Nic, I didn't know you were trying to quit smoking! Next time I'll send you my resume before I post, OK BOSS!

[kastro68]

It is nice to find humour every now and then.

[Nic]

LoL :D :D

Sorry...Didn't mean to be harsh :)

(I had/have to put up with so much of it in the Audio Encoding community that it does begin to grate after a while...but sorry to be harsh :)

-Nic

[ProfDrMorph]

I haven't seen a good wavelet implementation yet & although they are very promising in theory
Have you tried the latest version of LuraWave ( ACDSee-Plugin availible, www.acdsee.com ) or PicVideos Wavelet Codec ( "MotionJPEG2000" )? Those are already VERY good in my opinion. Sometimes if you're encoding an image to both, DCT (JPEG) and Wavelet, the Wavelet file might have less details but it's significantly smaller.

[sumpm1]

It's all good Nic, I understand.

But back to wavelets, it was hard enough for me to grasp dct with jpeg, I never went past it, because it takes such intense focus to understand some of the concepts. So, I am not even knowledged in video compression techniques(I know how they work) but not real technically. I have a decent understanding of jpeg compression.

[ProfDrMorph]

Before someone misunderstands me:

the Wavelet compressed images are so much smaller that raising quality gives you back the lacking details ( compared to the DCT file ) and the Wavelet is still smaller. I did not a lot of testing on different images ( only screenshots from UnrealTournament ) but red colored parts look MUCH better. I play a lot of UT CTF and so most of my screenshots contain team listings. The names of the people in the red team are sometimes hardly readable because of DCT compression ( I compress them later on to JPEG since my only usable Wavelet codec ( LuraWave ) is not widely spread and I only have a demonstration copy ).

[maven]

Originally posted by Koepi
Hm. My experience is that the wavelet codecs produce output that looks like shit....

Just my 2 € cent.

Regards,
Koepi

true. but that doesn't mean they're bad (but they aren't a miracle cure, either)...

if anyone is interesting, i did a wavelet coder for images you cann play with...
includes source, as always, feel free to give feedback.

http://www.maven.de -> code -> wavelet kompressor

but it'll be awhile before i come up with some sensible thought on MC (motion compensation).

[tangent]

Wavelet truely shines at the low bitrates, and is able to maintain more details than DCT codecs can at the low bpp without introducing blocking and halo/ringing artifacts. At higher bitrates, wavelets and DCT perform pretty much the same. When you see shitty wavelet results, it's because it's part of a demonstration to show that DCT will look shittier at the same bpp.

[tangent]

Ok, getting back to the original question, you won't likely see Wavelet in use that soon because much of it is still pretty much in research. It is not know the best method of using the DWT, and within DWT itself there are many variations of bases (Haar, Daubechies, Biorthogonal, etc) which produce varying results, some which are better for different uses of DWT. Even the Tarkin project itself has been through many discussions on how best to use wavelets. I've seen them talking about using wavelets for keyframes and MC+DCT for delta frames, and then talking about using wavelets for keyframes and MC+DWT for delta frames, and then talking about forgetting MC and using 3D DWT. I've no idea what they're working on now, really.

Even in audio, it can be pretty much agreed on that DWT cannot be used on it's own. Optimally, you want to split an audio signal into tonal and transient parts, encode the tonal part using MDCT and the transients using DWT. People claim to have worked on this and gotten good results, but I haven't seen any codecs yet. Ogg Vorbis is supposed to have this in version 2, but that's still a long time coming.

[yokem55]

My guess would be that wavelet support is probably a couple of mpeg-4 generations away.....B-frame support is just in its infancy (at least supposedly with divx5) at this point, and I would bet that the coders will be working on getting b-frames working well and well tweaked before moving on to some other new majore feature, such as wavelets....