Doing ffmpeg -codecs | grep -i huffyuv shows that my FFmpeg build supports three different versions of the HuffYUV lossless video codec:
DEVI.S ffvhuff Huffyuv FFmpeg variant
DEVI.S huffyuv HuffYUV
D.VI.S hymt HuffYUV MTI understand that HuffYUV MT is multi-threaded, and that my build only seems to support decoding it, but what are the differences between the standard HuffYUV codec and ffvhuff, FFmpeg's version of it? Is this documented anywhere?
2 Answers
On the ffmpeg encoding side of things there are two main differences.
Pixel formats
ffvhuff supports many more pixel formats (44 vs 3).
Refer to ffmpeg -h encoder=ffvhuff and ffmpeg -h encoder=huffyuv for the list of supported pixel formats for each encoder.
Per-frame Huffman tables
ffvhuff supports per-frame (context-adaptive) Huffman tables which can result in a smaller output file size. But it is slower to encode because it forces the thread count to 1.
Default is not to use per-frame Huffman tables. It can be enabled with -context 1.
Other than pixel formats and per-frame Huffman tables they are pretty much identical.
7What are the differences between the HuffYUV and ffvhuff codecs?
HuffYUV
HuffYUV was proposed by Ben Rudiak-Gould as an alternative to uncompressed Y CbCr video. It combines an intra-frame prediction with a consecutive Huffman entropy coding of the residuals. The intra-frame prediction selects between three different prediction models: left, gradient and median.
The first model, left, only uses the pixel l to predict the pixel x as x = l,
the second model, gradient, predicts x as x = l + a -d and
the median model selects the median value from the model left, the model gradient and from the pixel a above x
Note, that the model selection is not adaptive, but rather one model is selected a-priori to the encoding and then used for the complete video sequence. Also only a fixed table for the Huffmann code entropy encoding is used.
Added for context:
FFvYUV is an extension of HuffYUV developed by the FFmpeg project to address some of HuffYUV’s short-comings: instead of a fixed Huffman table, context-adaptive Huffman tables are used for the entropy encoding.
Source COMPARISON OF LOSSLESS VIDEO CODECS FOR CROWD-BASED QUALITY ASSESSMENT ON TABLETS (pdf) authored by Christian Keimel, Christopher Pangerl and Klaus Diepold.
The paper linked above includes a lot more information, images, and links to further resources.
ffvhuff
This codec was developed by ffmpeg.
It is a huffyuv like codec except it is not backwards compatible with huffyuv. However, it compresses better than huffyuv, supports yv12 colorspace, and is about as fast as huffyuv
Source ffvhuff? [Archive] - Doom9's Forum
6
