SRCNN
The PyTorch implementation of SRCNN for single image super resolution.
The following entry was written based on this repo.
https://buildersbox.corp-sansan.com/entry/2019/02/21/110000
Requirements
- torch
- torchvision
- pillow
- tensorboardx
- googledrivedownloader
Data preparation
You can use General-100 dataset through a script. Change your directory to ./data and run general100.py. It will download dataset from Google Drive and split the dataset into train/val/test randomly. The split ratio is (train, val, test) = (8, 1, 1).
$ cd ./data
$ python genaral100.py
Training
You can train the model through train.py.
$ python train.py
If you have a GPU, you should set the --cuda argument. It does not support multiple GPUs and data parallelization.
$ python train.py --cuda
Testing
Run test.py. You must specify the location of .pth file and a directory to save the results.
$ python test.py --weight_path ./runs/your/model/weight.pth --save_dir ./want/to/save/ --cuda