zzh-tech/ESTRNN — reverse-engineered prompt

Build me a Python project for video deblurring based on the ESTRNN research model. I want to be able to give it a blurry mp4 file or a folder of numbered video frames, choose a saved model checkpoint, and get a clean deblurred video or image sequence saved into an output folder.

Please include the main pieces needed to train and test the model on the BSD, GOPRO, and REDS datasets, with simple command line options for dataset path, dataset type, checkpoint path, batch size, learning rate, GPU count, and test only mode. It should support running on a GPU with PyTorch and include progress output so I can tell what is happening.

Make the setup clear for someone who is not an expert. Add a README with install steps, where to put checkpoints and datasets, example commands for training, testing, and deblurring my own video, plus notes about expected input frame names. Look up current docs online if you need to.