XVID 2-pass

[spyder]

Has anyone thought about making XVID's 2-pass work like DivX 5, creating a motion vector file during the first pass so motion searching doesn't have to be performed again? This really helps in the speed of the second pass on DivX 5. I hate DivX 5 because it is so slow using the new features and it's still slower and blurrier than DivX 4.12. I have almost completely switched to XviD because of the sharpness of the picture compared to DivX. I don't think it would be hard to implement but then again I'm not a C coder and don't know much about motion vectors.

[The People's Elbow]

I don't know, if the motion vector thingy has highest priority for the XviD developers atm, but I'm pretty sure they're already working on it or at least put it on the todo list!

PS: DivX5 blurrier than 4.12? Is that possible? ;)

- greetz from the Elbow

[spyder]

To me the post processing in DivX 5 is much more powerful than that of DivX 4. It seems to blur the entire image too much for me. It's also incredibly slow and sometimes jerky on my 1700+ Athlon XP.

[canadian_fbi]

i second the motion vector file idea. to me this was one of the most intriguing things in divx 5, as it gave a big speed boost simply by not computing something that's already been computed. most of divx 5's other gains seemed to come at some cost (doubling the quantizer for b-frames, for example), but this seems like something that would simply increase the second pass speed significantly at only the cost of the space of a small temp file.

[-h]

Caching the first-pass motion vectors isn't free from drawbacks - the motion vectors are no longer optimal, and quantizers are thus higher than they would be if the MV's were calculated from scratch again.

It wouldn't take long to hack in however. I might try it once this interlacing stuff works.

-h