2009/1/19 Stas Oskin <[email protected]>: > That's because the player hasn't decoded the all the reference frames >> for that frame to which you're trying to seek. It probably doesn't fix >> itself until it gets to the next I-frame. Having a larger GOP size >> should improve compression significantly in MPEG-4 Part 2. P-/B-frames >> are much smaller than I-frames and having larger GOPs means less >> I-frames at the cost of seekability and error resilience. It's your >> choice. > > > Thanks for the explanation, I actually figured out almost same myself. > > My question is, how GOP can improve the quality, and not the compression? In > the original answer you said that larger GOP increases quality - but if the > distance between key-frames increases, then actually the decoder needs to > guess more, and the quality drops?
Improving compression efficiency is mostly equivalent to improving quality at a particular bit rate. I-frames are completely coded. P-/B-frames are predicted from other frames. That is, the frame data contains information to reconstruct motion vectors that point to positions in previous (or future in some cases) frames and the difference between this prediction and what should actually be seen. The pixel data from these reference frames is combined with the difference to produce the desired result. It's not really guessing, rather minimising the difference between the prediction and the source. Using predicted frames (P-/B-frames) improves compression efficiency (and so quality at a particular bit rate) significantly because there is a lot of temporal redundancy in video. Frames are similar to previous/future frames in close proximity temporally. Predicting a frame from those in close proximity temporally allows for more information per bit, which is what compression efficiency is really. That's how most video codecs work. Regards, Rob _______________________________________________ libav-user mailing list [email protected] https://lists.mplayerhq.hu/mailman/listinfo/libav-user
