README.md

Training

1. Download restructured RealSR version 2 for training:

python download.py --file RealSR_234

2. (Optional) To download generated dataset with intermidiate levels from Interflow:

python download.py --file RealSR_24_gen

3. To train SR networks, run

python3 main.py \
--GT-paths [TRAIN DATASET PATH] \
--parameter-name [FOLDER NAME TO SAVE THE RESULTS] \
--train-iteration [NUM TRAIN ITERATION] \
--validation-step [VAL ITERATION] \
--test-scale 3.0 \
--model [VDSR|RCAN|KPN|HAN|NLSN|SwinIR|MetaSR|LIIF] \
--gpu [GPU INDEX];

Please refer more examples in scripts/train.sh

Evaluation

1. Download the pre-trained model and place it in ./pretrained_models/ manually

OR run:

python download.py --file pretrained

2. To download RealSR x3 test datasets[link], run

python download.py --file RealSR_234

3. To download DRealSR x3 test datasets [offical link], run

python download.py --file DRealSR_3_test

4. Testing

python3 main.py \
--parameter-name [FOLDER NAME TO SAVE THE RESULTS]  \
--parameter-restore-path [PATH TO YOUR CHECKPOINT] \
--test-scale 3.0 \
--model [VDSR|RCAN|KPN|HAN|NLSN|SwinIR|MetaSR|LIIF] \
--test-data [RealSR|DRealSR] \
--gpu [GPU INDEX] \
--test; 

Please refer more examples in scripts/test.sh