[FFmpeg-devel] video filter with R8G8B8 input and R10G10B10 output

[Guo, Yejun]

Hi,

I'm trying a deep learning model with converts SDR (with format R8G8B8) to HDR (with format R10G10B10) with ffmpeg. I wrote a video filter with only supported format AV_PIX_FMT_RGB24 (see code below), it turns out that the format of the input frame of the filter is R8G8B8, and so the output frame, the input and output formats are always the same.

my question is how to write a filter with AV_PIX_FMT_RGB24 as input format, and with AV_PIX_FMT_GBRP10BE/AV_PIX_FMT_GBRP10LE as output format.

// here is how I currently set the format for the filter.
static int query_formats(AVFilterContext* context)
{
    printf("%s:%d query_formats called\n", __FILE__, __LINE__);
    const enum AVPixelFormat pixel_formats[] = {AV_PIX_FMT_RGB24,
                                                AV_PIX_FMT_NONE};
    AVFilterFormats* formats_list;
    formats_list = ff_make_format_list(pixel_formats);
    if (!formats_list){
        av_log(context, AV_LOG_ERROR, "could not create formats list\n");
        return AVERROR(ENOMEM);
    }
    return ff_set_common_formats(context, formats_list);
}

thanks
Yejun