On Mon, Apr 18, 2011 at 10:09:01AM -0400, Ronald S. Bultje wrote:
> Hi,
> 
> I'd like some eyes on this patch and it's visual effects.
> 
> To test, please try "ffplay tests/data/vsynth1/error-mpeg4-adv.avi"
> before and after applying the patch. Also see:
> http://ronald.bitfreak.net/images/er.png
> Where left is always after the patch and right is before. Some people
> on IRC already noted: some parts are worse (blue/red rectangles),
> other parts (especially the splotch of diagonal lines in the
> bottom-right) look better after the patch. I personally feel that the
> ffplay-sequence looks better to the eye after the patch, but would
> like opinions. You can try vsynth2-error also but the MVs are pretty
> constant there so you won't see much of a difference.
> 
> What the patch does:
> - current code uses current_picture.motion_val[cur_idx] to access the
> previous frame's MVs. This is wrong. current_picture doesn't hold the
> previous frame's MVs, they are the MVs of several frames ago for which
> wer'e reusing the same memory pointer. The last frame is in
> last_frame.motion_val[]. The reason this is non-deterministic is
> because the value of the memory pointer differs if threading is used.
> For last_frame.*[], this becomes deterministic.
> - so the patch changes it to use last_frame instead of current_frame.
> 
> Please let me know opinions.

Sounds reasonable, looks reasonable and produces equal output in terms of
quality too, so OK for me. 
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to