README.md

Training

Training on LoL dataset

• Download LoL training and testing data, run

python download_data.py --data train-test --dataset Lol

• To train MIRNet_v2, run

cd MIRNetv2
./train.sh Enhancement/Options/Enhancement_MIRNet_v2_Lol.yml

Training on MIT-Adobe Fivek dataset

• For MIT-Adobe Fivek training data, download DNGs from https://data.csail.mit.edu/graphics/fivek/ and then follow this for data preparation nothinglo/Deep-Photo-Enhancer#38

• Download Fivek mini validation data, run

python download_data.py --data val --dataset FiveK

• Generate image patches from full-resolution training images

python generate_patches_fivek.py 

• To train MIRNet_v2, run

cd MIRNetv2
./train.sh Enhancement/Options/Enhancement_MIRNet_v2_FiveK.yml

Note: The above training script uses 8 GPUs by default. To use any other number of GPUs, modify Restormer/train.sh and Enhancement/Options/Enhancement_MIRNet_v2_FiveK.yml

Evaluation

Testing on LoL dataset

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

wget https://github.com/swz30/MIRNetv2/releases/download/v1.0.0/enhancement_lol.pth -P pretrained_models/

• Download LoL testset, run

python download_data.py --data test --dataset Lol

• Testing

python test.py --dataset Lol

Testing on MIT-Adobe Fivek dataset

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

wget https://github.com/swz30/MIRNetv2/releases/download/v1.0.0/enhancement_fivek.pth -P pretrained_models/

• Download MIT-Adobe Fivek testset, run

python download_data.py --data test --dataset FiveK

• Testing

python test.py --dataset FiveK

To reproduce PSNR/SSIM scores of the paper (Table 5) on LoL, run this MATLAB script

evaluate_PSNR_SSIM.m