Increase the quality of your movies using a superres deep neural network inference on your GPU.

For installation, you need to have openCV compiled with dnn_superres module.

git clone https://github.com/thomasarmel/movie_quality_increase.git
cd movie_quality_increase
cmake .
make

./movie_quality_increase -f <upscale factor (2 or 4)> -i <input file path> -o <output file path> -m <models directory path>

Upscale factor: The upscale factor to use, only 2 and 4 are supported by the ESPCN model used here.

Input file path: The path to the input video file, must be readable by ffmpeg library.

Output file path: The path to the output video file, must be writable by ffmpeg library. Note that the output file will be overwritten if it already exists.

Models directory path: The path to the directory containing the models, provided in the repository.

We will upscale per 2 the movie old-movie.mp4 to new-movie.mp4 using the ESPCN model.

./movie_quality_increase -f 2 -i old-movie.mp4 -o /tmp/upscale-only-video.mp4 -m ./models
ffmpeg -i old-movie.mp4 -i /tmp/upscale-only-video.mp4 -c copy -map 0:a? -map 1:v -map 0:s? -shortest new-movie.mp4

Thanks to @fannymonori and @Saafke for providing superres models.