On Thu, Feb 09, 2017 at 09:39:13 -0500, Chao Zhou wrote: > > I expect the size of total 9 segments will be larger than the source video > > When encoding source video, the P/B frame can search previous full size I > frame to reduce their encoding size. After cropping, each of these video > segments has a smaller size of I frame. Then, the following P/B frame have > to encode more redundant data for correct recovery on the decoder side. Is > it right? I didn't change any bitrate or video quality parameters during > cropping process.
You did change them, because you decode the original video and then reencode it with ffmpeg/libx264. The resulting bitrate depends on various parameters, which are not related to the original video: encoding profile and level, CRF value, various other encoding parameters, and last not least encoding efficiency of the utilized encoder. You can probably vary the bitrate by about a factor of 10x with little variation in the visual output quality. (YMMV) If you use a different H.264 encoder, you will most likely get different results as well! To get back to your original question as to why your observed relation no longer applied to "larger scale" videos: Perhaps the originals were *less* efficiently encoded, so the sum of your segment encodings became relatively smaller. Or they were different material (less motion, animation, or so) leading to better encoding efficiency. Or - long shot, and I doubt it - something else made the encoder more efficient at the higher resolutions. > [libx264 @ 0x332d380] frame I:4 Avg QP:18.31 size: 33702 > [libx264 @ 0x332d380] frame P:261 Avg QP:20.60 size: 3323 > [libx264 @ 0x332d380] frame B:709 Avg QP:25.26 size: 268 > [libx264 @ 0x332d380] consecutive B-frames: 2.8% 0.4% 0.3% 96.5% You might want to compare some of these statistics. On the other hand, they may be misleading - I don't really know. ;-) Moritz _______________________________________________ ffmpeg-user mailing list [email protected] http://ffmpeg.org/mailman/listinfo/ffmpeg-user To unsubscribe, visit link above, or email [email protected] with subject "unsubscribe".
