Implementation of SRCNN in PyTorch.
To train the model with a zoom factor of 2, for 200 epochs and on GPU:
python main.py --zoom_factor 2 --nb_epoch 200 --cuda
At each epoch, a .pth model file will be saved.
To use the model on an image: (the zoom factor must be the same the one used to train the model)
python run.py --zoom_factor 2 --model model_199.pth --image example.jpg --cuda
don't do this unless you have to
docker build . -t srcnn
docker run --gpus all -it -v $PWD:/data srcnn bash
python scale.py -if 0 -lf 750 -ip /input_frames/dada/steps -sp /scaled_frames/dada -op /output/dada.mp4 -zf 2 -fps 10
Original image:
Bicubic interpolation zoom:
SRCNN zoom:
Original paper on SRCNN by Dong et al. (Image Super-Resolution Using Deep Convolutional Networks)