There is a slight color shift on video inferencing

I believe it has to do with YCbCr to/from RGB conversion. The RGB image inferencing is not affected

To test, I changed in the inference_video_plus_...py scripts the cv2.VideoWriter_fourcc to FFVH for lossless compression, and it also supports many pixel types, but the output is always YUV420P8.

You can also see this more easily on greyscale video input. Instead of the input Cb=Cr=128 , the output video has deviations between 125-130 and a tint appears even on non interpolated frames.

If you need any assets or examples to clarify, please let me know

Thanks for developing this interesting project