On Tue, Jun 10, 2014 at 12:20:44PM +0200, Maxim Polijakowski wrote: > Am 08.06.2014 14:15, schrieb Dirk Ausserhaus: > >On Sun, Jun 8, 2014 at 2:10 PM, Kostya Shishkov > ><[email protected]> wrote: > >>On Sun, Jun 08, 2014 at 01:53:49PM +0200, Dirk Ausserhaus wrote: > >>>Here's a patch that fixes decoding of Indeo 4 B-frames. > >>Looks correct. I should've done it long time ago and saved you some work. > >> > >>>Now the main problem is Indeo 4 B-frames reconstruction. For that I > >>>need to modify IVI context to store the second pair of vectors (that > >>>is not hard at all) and make it perform the averaging motion > >>>compensation for B-blocks (that one seems to require some more > >>>hacking). > >>It seems tricker than most codecs since in Indeo one does not add residue to > >>motion compensation but rather the othe way around so motion function will > >>have to do averaging inside before adding result to the block. > > IIRC, it performs a simple averaging motion compensation like this: > > (X + Y) / 2. > > IIRC, it doesn't even do a proper rounding.
Should not matter much in this case. > The tricky part was which frames to choose - two different kinds of > NULL frames (type 6 and 7) play the key role there. > I'll try to search my data for more information on this topic. Not > sure how sucessful it will be because alot of time has passed since > I worked on this codec... At least in volcano.avi you have a sequence of I, I, I, IP, B, B, type6 (null_last), type4 (droppable P), IP, B, B, type6 Another file I have features this pattern: I, I, I, IP, B, B, P, B, B, P Hopefully Dirk will figure something out. _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
