Hi I’m new to using Natron. Can anyone tell me, for sure, what chroma resampling modality (Bilinear, Bicubic, Nearest-Neighbor/Point etc) Natron uses (by default) for sub-sampling RGB/444 to (Rec709) YUV 422 at export -that is to say, when there is no change in resolution, only format ?
Also, is there any setting that allows choice of the resampling modality ? Can’t find anything in the Natron User Guide.
Thanks.
ffmpeg does it, so see the ffmpeg documentation
FFmpeg uses its own swscale library for both resizing and pixel format conversions. According to the documentation (linked in previous sentence), it should default to bicubic interpolation.
… However, if you actually use FFmpeg, you’ll find that is not accurate. If you’re actually using the scale filter and specify the flags filter parameter without actually setting a scaling algorithm, yes it defaults to bicubic - but if you leave out the flags parameter entirely, it uses bilinear instead.
At least, it used to. I’m not 100% sure if it still does, but I felt I should chime in just in case. I’m also not sure if this behavior is the same when you use the APIs more directly (rather than calling the ffmpeg command-line application, which is where most of my experience with FFmpeg is).
Thanks guys. Actually I figured it out using a two-color 5X5 pixel/block checkerboard image (created in Natron) exported to ProRes_4444 and ProRes_HQ (10bit 422). I then compared the block patterns and scope profiles when the ProRes_4444 file was converted to Uncompressed 10bit 422 (v210) with AVISynth and Vaporsynth using different chroma resamplers. The results were consistent with simple Box kernel sub-sampling.