On Mon, 11 Jun 2007, Yannick Patois wrote: > Most of the texture is lost in all rendered images... > > Thanks for any suggestion,
You seem to be forcing a large GOP size with -g 15 -G 15. That means it's only allowed to emit an I-frame (complete description of a frame) twice per second, and everything else has to be compressed by referring to changes from the I-frames. Since this video contains rapid full-screen motion, the image is totally different from one I-frame to the next, and so you can't compactly describe most of the frames in between in terms of differences from the I-frames; the compressor has to re-describe each frame almost completely anyway. I'd suggest setting a much smaller minimum GOP size, like "-g 5", with a maximum bigger than the minimum, so that the compressor can choose GOP sizes instead of being forced to have an I-frame exactly once every 15 frames. Then it can make a better attempt at conserving frame-to-frame information. Two-pass encoding would probably also help, if you're not already doing it. I also wonder if your requirement that the texture be preserved is really reasonable. You've got a complicated pseudorandom texture (looks like Perlin noise, for instance from POV-Ray), which won't compress well at the best of times, and you're trying to change it pretty much completely every single frame. Information theoretic considerations mean that it's just not possible to really do that well, no matter what algorithm or settings you use. You are trying to fit far more than 8600kbps of entropy through an 8600kbps pipe. That's an impossible feat. MPEG compression is dependant on faking it, with a psychovisual model of what human beings can actually see. It preserves the information that is really visible, throws away the rest, and the quality of the output is determined by how well the encoder guesses as to what's visible. It seems like the model is predicting that under these conditions the viewer won't really be able to see your texture, and the thing is, I think the model may actually be *right*. Maybe it would be more clear in a longer sample, but in your one-second clip, I can't focus on the white part firmly enough to be able to see a texture if one were there - it's moving so fast as to be a blur anyway. Can you really see the difference in quality between the compressed and uncompressed versions under actual viewing conditions at full speed? It's easy to spot the difference in still frame grabs, but MPEG cannot, and doesn't attempt to, preserve all details that might be visible with frame grabs. It's meant for moving video. -- Matthew Skala [EMAIL PROTECTED] Embrace and defend. http://ansuz.sooke.bc.ca/ _______________________________________________ Cinelerra mailing list [email protected] https://init.linpro.no/mailman/skolelinux.no/listinfo/cinelerra
