MJPEG-2000 With IP & MC?

[CyberGuy]

Has anybody played with Motion JPEG-2000 with interframe prediction and motion compensation? While it seems to be comparable to MPEG 4.10 AVC/H.264, it has the benefit of scaling very well. I encoded the first 654 frames of Elephants Dream using lossless compression and was able to display all 654 frames in the 850MB file in less than two seconds at a reduced resolution. I did not use interframe prediction or motion compensation.

The only software that I was able to find that did MJPEG-2000 with interframe prediction and motion compensation was MC-EZBC, but it is not optimized and buggy. I’m interested in playing with D-Cinema like I did with the JM Reference software when it first appeared. Is there any software available to test that uses IP and MC. I believe that SNOW is similar, but probably doesn’t conform to the MPJEG-2000 standard.

[benwaggoner]

The big challenge for wavelet codecs in the video domain is how to integrate motion compensation efficiently. While wavelets have a ton of promise for still images, we're still waiting for a compelling example of a wavelet-based interframe codec. There DCT and its derivatives have long ruled the roost.

[akupenguin]

You seem to be confusing "MJPEG-2000" with "wavelet". MJPEG-2000 is just a sequence of JPEG-2000 images, like MJPEG is a sequence of JPEG images. There is no interframe prediction in either one. MC-EZBC is something else entirely, and the only thing it has in common with JPEG-2000 is that both use wavelets. Same with Snow, Dirac, and 3D-SPIHT.

[Dr Pizza]

Is there much software support for MJPEG2000? It's something I've been meaning to investigate, as it's something that could be professionally useful to me (as an archival format).

[benwaggoner]

Is there much software support for MJPEG2000? It's something I've been meaning to investigate, as it's something that could be professionally useful to me (as an archival format).
The only significant use I've seen for it for video is in the DCI digital film standard.

That said, the intense decode complexity required to do a 2K J2K decode every 1/24th of a second has been a pretty big issue.

Simpler formats like Cineform also provide >8-bit encoding with much better performance.

[Dr Pizza]

Our interest is in MJPEG2000 as a format that's reasonably well-documented, usable in both lossy and lossless modes, and editable. Playback performance isn't so important (we can always transcode for access to material). The only thing I'm not sure about is its handling of interlacing. It would be preferable not to have to deinterlace interlaced material.

[benwaggoner]

Our interest is in MJPEG2000 as a format that's reasonably well-documented, usable in both lossy and lossless modes, and editable. Playback performance isn't so important (we can always transcode for access to material). The only thing I'm not sure about is its handling of interlacing. It would be preferable not to have to deinterlace interlaced material.
If it's MJPEG, it'll definite have interlaced (the main features of the original MJPEG were interlaced at 4:2:2).

That said, editable would require good performance in both encode and decode. HD for real-time JPEG2000 is right out, and even SD would be a serious strain - like DV editing in sofwtare back in '97.

[foxyshadis]

The same companies that make MJPEG offload and capture cards also make MJPEG2000 cards, if you need instant HD performance. SD editing is actually not onerous at all, realtime play is no problem, though Morgan's codec is single-threaded and thus entirely unsuitable for HD. I will admit that MJ2K definitely pushes the performance down as bitrate increases, becoming as painful to use as long-gop divx when you get around 100mbps on mine, presumably mainly because of the arithmetic encoding.

[Dr Pizza]

If it's MJPEG, it'll definite have interlaced (the main features of the original MJPEG were interlaced at 4:2:2).

That said, editable would require good performance in both encode and decode. HD for real-time JPEG2000 is right out, and even SD would be a serious strain - like DV editing in sofwtare back in '97.
Ah, but you must remember that I'm in this for the long haul. I don't need a format I can play on contemporary hardware--I am happy to generate MPEG2/WMV/QT/Real/whatever else is convenient at the time for playback. I want a format that gives me the option of lossless encoding, is editable, is HD-capable, and is published. I'm not even too bothered about patents--we can afford to wait them out.

[akupenguin]

Then what is your definition of "editable"? If you don't care about cpu usage, then any lossless format will work.

[Dr Pizza]

Any format in which a random frame can be decoded independently of any other frame.