Skip to content

Latest commit

 

History

History
50 lines (40 loc) · 1.54 KB

README.md

File metadata and controls

50 lines (40 loc) · 1.54 KB
Raw

Training

  1. Download training data (RealSR version 1), run
python download_data.py --data train
  1. Generate image patches from full-resolution training images, run
python generate_patches.py --scale x2
python generate_patches.py --scale x3
python generate_patches.py --scale x4
  1. To train MIRNet_v2 with default settings, run
cd MIRNetv2
./train.sh Super_Resolution/Options/SuperResolution_MIRNet_v2_scale2.yml
./train.sh Super_Resolution/Options/SuperResolution_MIRNet_v2_scale3.yml
./train.sh Super_Resolution/Options/SuperResolution_MIRNet_v2_scale4.yml

Note: The above training script uses 8 GPUs by default. To use any other number of GPUs, modify Restormer/train.sh and the yaml file correspondng to each SR scaling factor (e.g., Super_Resolution/Options/SuperResolution_MIRNet_v2_scale2.yml)

Evaluation

  • Download the pre-trained models and place them in ./pretrained_models/:
wget https://github.com/swz30/MIRNetv2/releases/download/v1.0.0/sr_x2.pth -P pretrained_models/
wget https://github.com/swz30/MIRNetv2/releases/download/v1.0.0/sr_x3.pth -P pretrained_models/
wget https://github.com/swz30/MIRNetv2/releases/download/v1.0.0/sr_x4.pth -P pretrained_models/
  • Download test datasets (for x2, x3, x4 scale factors), run
python download_data.py --data test
  • Testing
python test.py --scale x2
python test.py --scale x3
python test.py --scale x4

To reproduce PSNR/SSIM scores of Table 4, run

evaluate_PSNR_SSIM.m