Loren Merritt schrieb: > On Fri, 14 Mar 2008, Thorsten Jordan wrote: > >> What about h264, that choses MBAFF on its own, which format must the YUV >> data have for that codec, always progressive or always interlaced? What >> happens on MBAFF/PAFF? >> Can some of the h264 experts give me a hint here please? thanks > > I don't know. But I can only think of 2 possibilities that make sense. > A) Specify chroma subsampling format somewhere in the VUI along with all > the other colorspace metadata. If that is the case, then it's > user-specifiable but not specified by default in x264, and independent > of how any given frame or block is coded. > B) Assume interlaced chroma if any interlace-related features are enabled > anywhere in the stream, otherwise assume progressive. > > Currently I think mplayer uses (B), but (A) is more likely to be the > standard. well ffh264 sets the interlaced_frame flag when the frame is MBAFF or PAFF, and that flag is also used for MPEG2 to distuingish between interlaced and progressive color coding - at least that method was valid so far, so rather B) is used, yes
>> libx264.c doesn't use the INTERLACED flags mentioned above, so i can't >> tell the encoder what to use... > > x264 doesn't *care* what the chroma format is, it just encodes samples. > The default values x264 puts in the VUI are "unspecified", so if you want > to specify some metadata for the display to use (assuming the decoder also > preserves that info, which ffh264 doesn't), then figure out which vui > element and use that option (which is available in libx264, but not in > lavc-libx264). ah I see, the UV data is just handled as block of color information, similar like Y, the encoder/decoder pair tries to replicate the original as close as possible - that is indeed independent of the interlaced/progressive color coding. I can only think that it would make a tiny difference in compression efficiency, as there may be color combs in the data, which are harder to encode... Thanks for this valuable information, Loren! -- Regards, Thorsten _______________________________________________ libav-user mailing list [email protected] https://lists.mplayerhq.hu/mailman/listinfo/libav-user
