README.md

Train

Prepare training data in data/train directory as below:

  data
  └── train
      ├── video_1
            ├── hr
                    ├── hr0.png
                    ├── ...
                    └── hr30.png
            └── lr_x4_BI
                    ├── lr0.png
                    ├── ...
                    └── lr30.png
      ├── ...
      └── video_N

• Run on CPU:

python train.py --scale 4 --patch_size 32 --batch_size 32 --n_iters 200000

• Run on GPU:

python train.py --scale 4 --patch_size 32 --batch_size 32 --n_iters 200000 --gpu_mode True

Test

We provide the pretrained models (2x/3x/4x SR on BI degradation model and 4x SR on BD degradation model) for evaluation on the Vid4 dataset.

• Generate LR test images (Matlab)

  • Run data/test/generate_LR_images.m

• Inference

  • Run on CPU:

   python demo_Vid4.py --degradation BI --scale 4

  • Run on GPU:

   python demo_Vid4.py --degradation BI --scale 4 --gpu_mode True

  • Run on GPU (memory efficient):

   python demo_Vid4.py --degradation BI --scale 4 --gpu_mode True --chop_forward True

• Evaluation (Matlab)

  • Run evaluation.m