Snow looks interesting. It appears to be part of ffmeg. But I have found it very difficult to get any further information about it. Anyone know more about Snow?
The posting you mention also mentions Rududu, which also looks interesting but doesn't seem to be open source (the download is a binary only).
Comments and more info most welcome.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
a quote from a recent discution :
--------
> From google all I've found out is
> that it's an experimental wavelet codec written by the ffmpeg team. What
> kind of wavelet transform does it use? Daubechies? Symmetric? Does it
a symmetric biorthoginal integer approxmation of the 9/7 daubechies wavelet
> use motion vectors?
yes
> 3D wavelets?
no, normal 3d wavelets are simply a very bad choice for video compression,
there are other lifting based temporal transforms which do work well, but
they add some delay, higher memory requirment and they would slow snow down
further
> Are there any license/patent issues?
iam not aware of any patents which cover it except one from IBM which IIRC
will expire in a few month if it didnt already
if you want a certainly patent free codec you have no choice except using one
which is older then the expiration timespan for patents
anyone who claims that a codec which is less old to be patent free is lying
the number of patents is so huge that checking them all is simply not
possible
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
FWIW I've just uploaded a mod so that Dirac can use a range of wavelet transforms, including the approximate Daubechies (9,7) one. As far as I can see from the snow code, the choice of wavelet filter is a compile-time selection.
I don't know what the patent position with snow is, because there's no documentation of the algorithm, so it's hard to tell where it might infringe prior art. As far as I can tell from the code, it's similar in overall structure to Dirac, but less flexible in terms of chroma formats, block sizes and temporal prediction formats. We've not done any tests to compare quality, though. Perhaps someone can have a go?
Thomas
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I found this discussion in the Doom9 forum:
http://forum.doom9.org/showthread.php?s=3aab15fb60a8fb6c0b362accc3f906fd&threadid=84593&highlight=snow
It's about the "snow" wavelet codec. Seem it is almost ready and has good performance.
Perhaps this can be of help?
Thanks for the posting.
Snow looks interesting. It appears to be part of ffmeg. But I have found it very difficult to get any further information about it. Anyone know more about Snow?
The posting you mention also mentions Rududu, which also looks interesting but doesn't seem to be open source (the download is a binary only).
Comments and more info most welcome.
Post a mail to ffmpeg ml to have futher info.
a quote from a recent discution :
--------
> From google all I've found out is
> that it's an experimental wavelet codec written by the ffmpeg team. What
> kind of wavelet transform does it use? Daubechies? Symmetric? Does it
a symmetric biorthoginal integer approxmation of the 9/7 daubechies wavelet
> use motion vectors?
yes
> 3D wavelets?
no, normal 3d wavelets are simply a very bad choice for video compression,
there are other lifting based temporal transforms which do work well, but
they add some delay, higher memory requirment and they would slow snow down
further
> Are there any license/patent issues?
iam not aware of any patents which cover it except one from IBM which IIRC
will expire in a few month if it didnt already
if you want a certainly patent free codec you have no choice except using one
which is older then the expiration timespan for patents
anyone who claims that a codec which is less old to be patent free is lying
the number of patents is so huge that checking them all is simply not
possible
FWIW I've just uploaded a mod so that Dirac can use a range of wavelet transforms, including the approximate Daubechies (9,7) one. As far as I can see from the snow code, the choice of wavelet filter is a compile-time selection.
I don't know what the patent position with snow is, because there's no documentation of the algorithm, so it's hard to tell where it might infringe prior art. As far as I can tell from the code, it's similar in overall structure to Dirac, but less flexible in terms of chroma formats, block sizes and temporal prediction formats. We've not done any tests to compare quality, though. Perhaps someone can have a go?
Thomas