About

Fast CUDA FFMPEG deinterlacer. Note that requires specific CUDA driver versions. Works on DiGiTS-Dev-Box-Fish as of 9-18-20

Credits

Many thanks to https://github.com/Svechnikov for his example Dockerfile.

Requirements

• NVIDIA graphics card
• CUDA Driver Version: 440.100, Version: 10.1
• Docker

Build for CUDA 10.1

docker build -t mbari/cuda-deinterlace .

Examples

Convert prores to deinterlaced compressed h264

• -map_metadata 0:g Copies global metadata
• -c:v h264_nvenc Use CUDA enabled h264 encoder
• -vf yadif=1 De-interlace and convert to progressive scan
• -vf format=yuv420p Drop 10-bit encoding (not supported)
• -an Drop the audio
• -movflags faststart Move the mov atom to the start of the file for fast playback. Happens at the end of transcoding in second pass
• CUDA_VISIBLE_DEVICES=0 Run on device 0

docker run -e CUDA_VISIBLE_DEVICES=0 -it -u $(id -u):$(id -g) \
	-v $PWD:/io -v /mnt/M3/:/mnt/M3 --rm mbari/cuda-deinterlace \
	ffmpeg -y -hwaccel cuvid -i /mnt/M3/master/DocRicketts/2016/11/0904/D0904_20161113T020253Z_prores.mov \
        -map_metadata 0:g -c:v h264_nvenc -preset slow -movflags faststart \
	-vf bwdif=0,format=yuv420p -movflags faststart \
	/io/bwdifD0569_20131213T224337Z_00-00-01-00TC_h264.mp4

docker run -e CUDA_VISIBLE_DEVICES=0 -it -u $(id -u):$(id -g) \
	-v $PWD:/io --rm mbari/cuda-deinterlace \
	ffmpeg -y -hwaccel cuvid -i /io/D0569_20131213T224337Z_00-00-01-00TC_prores.mov \
        -map_metadata 0:g -c:v h264_nvenc -preset slow -movflags faststart \
	-vf bwdif=0,format=yuv420p -movflags faststart \
	/io/bwdifD0569_20131213T224337Z_00-00-01-00TC_h264.mp4

Convert prores to deinterlaced prores using the bwdif filter. Prores does not currently have encode/decode CUDA support

docker run -it -v $PWD:/io --rm mbari/cuda-deinterlace  -u $(id -u):$(id -g) \
	ffmpeg -y -i /io/D0569_20131213T224337Z_00-00-01-00TC_prores.mov \
        -map_metadata 0:g -c:v prores -vf bwdif=0,format=yuv420p -movflags faststart \
	/io/bwdifD0569_20131213T224337Z_00-00-01-00TC_h264.mov