Repository navigation
Conversation
|
As far as I know, the alpha mode field is only set by ffmpeg for the JPEG-XL codec, not for anything we use. If we set it like proposed in the PR, there could even be a mismatch between the value coming from ffmpeg and the property in the HEIF layer. For aspect ratio, this may be set for H.265, but not for AV1. And we will also have potential conflicts with the HEIF layer. I would prefer not to read these fields from ffmpeg. |
|
|
|
For aspect ratio, the code would have to check at least for the undefined case (num > 0 && den > 0). |
|
BTW: If we read the SAR from H.265, we should also write it in encoders. See: a26561d |
…1860) Write the image's pixel aspect ratio into the codec-level signalling where the format supports it, so that the bitstream agrees with the pasp property: - x265, kvazaar (HEVC), x264 (AVC), uvg266, vvenc (VVC): Extended_SAR in the SPS VUI - JPEG: JFIF APP0 density fields with unit 0 (Xdensity:Ydensity = hSpacing:vSpacing) Only ratios that fit the u(16) VUI fields are written, 1:1 is omitted, and auxiliary images (alpha, depth) are excluded. AV1 and raw JPEG 2000 codestreams have no aspect ratio signalling and rely on the pasp property alone.
|
I think it's always necessary to encode and decode metadata in the bitstream by codec plugins whenever it is possible, let the HEIF core decide what to do with it. I'm closing this one, it seems this isn't necessary anymore. |
No description provided.