Install required libraries.
pip install -r requirements.txt
Train IMDN model by following command.
- root: the training data directory
- scale: scale factor
- nEpochs: total epochs
- step_size: learning rate descreased every step epoch
python train_IMDN.py --root train_decoded/ --scale 3 -nEpochs 200 --step_size 50
The pretrained model is in checkpoint_x3. You clone the whole github to inference.
You can reproduce the results by the following command.
- test_lr_folder: testing data directory
- output_folder: output directory
- checkpoint: model weights
- upscale_factor: upscale faactor
python inference.py \
--test_lr_folder testing_lr_images/ \
--output_folder results_30epoch \
--checkpoint checkpoint_x3/epoch_30.pth \
--upscale_factor 3
- IMDN: https://github.com/Zheng222/IMDN
- IMDN paper: https://arxiv.org/pdf/1909.11856v1.pdf
- VDSR: https://github.com/twtygqyy/pytorch-vdsr
- Seven ways to improve example-based single image super resolution, https://arxiv.org/abs/1511.02228