Skip to content

Latest commit

 

History

History
176 lines (134 loc) · 5.17 KB

README.md

File metadata and controls

176 lines (134 loc) · 5.17 KB
Raw

Image Denoising

Gaussian Image Denoising

  • Blind Denoising: One model to handle various noise levels
  • Non-Blind Denoising: Separate models for each noise level

Training

  • Download training (DIV2K, Flickr2K, WED, BSD) and testing datasets, run
python download_data.py --data train-test --noise gaussian
  • Generate image patches from full-resolution training images, run
python generate_patches_dfwb.py
  • Train Restormer for grayscale blind image denoising, run
cd Restormer
./train.sh Denoising/Options/GaussianGrayDenoising_Restormer.yml
  • Train Restormer for grayscale non-blind image denoising, run
cd Restormer
./train.sh Denoising/Options/GaussianGrayDenoising_RestormerSigma15.yml
./train.sh Denoising/Options/GaussianGrayDenoising_RestormerSigma25.yml
./train.sh Denoising/Options/GaussianGrayDenoising_RestormerSigma50.yml
  • Train Restormer for color blind image denoising, run
cd Restormer
./train.sh Denoising/Options/GaussianColorDenoising_Restormer.yml
  • Train Restormer for color non-blind image denoising, run
cd Restormer
./train.sh Denoising/Options/GaussianColorDenoising_RestormerSigma15.yml
./train.sh Denoising/Options/GaussianColorDenoising_RestormerSigma25.yml
./train.sh Denoising/Options/GaussianColorDenoising_RestormerSigma50.yml

Note: The above training scripts use 8 GPUs by default. To use any other number of GPUs, modify Restormer/train.sh and the yaml file corresponding to each task (e.g., Denoising/Options/GaussianGrayDenoising_Restormer.yml)

Evaluation

  • Download the pre-trained models and place them in ./pretrained_models/

  • Download testsets (Set12, BSD68, CBSD68, Kodak, McMaster, Urban100), run

python download_data.py --data test --noise gaussian

Grayscale blind image denoising testing

  • To obtain denoised predictions, run
python test_gaussian_gray_denoising.py --model_type blind --sigmas 15,25,50
  • To reproduce PSNR Table 4 (top super-row), run
python evaluate_gaussian_gray_denoising.py --model_type blind --sigmas 15,25,50

Grayscale non-blind image denoising testing

  • To obtain denoised predictions, run
python test_gaussian_gray_denoising.py --model_type non_blind --sigmas 15,25,50
  • To reproduce PSNR Table 4 (bottom super-row), run
python evaluate_gaussian_gray_denoising.py --model_type non_blind --sigmas 15,25,50

Color blind image denoising testing

  • To obtain denoised predictions, run
python test_gaussian_color_denoising.py --model_type blind --sigmas 15,25,50
  • To reproduce PSNR Table 5 (top super-row), run
python evaluate_gaussian_color_denoising.py --model_type blind --sigmas 15,25,50

Color non-blind image denoising testing

  • To obtain denoised predictions, run
python test_gaussian_color_denoising.py --model_type non_blind --sigmas 15,25,50
  • To reproduce PSNR Table 5 (bottom super-row), run
python evaluate_gaussian_color_denoising.py --model_type non_blind --sigmas 15,25,50

Real Image Denoising

Training

  • Download SIDD training data, run
python download_data.py --data train --noise real
  • Generate image patches from full-resolution training images, run
python generate_patches_sidd.py
  • Train Restormer
cd Restormer
./train.sh Denoising/Options/RealDenoising_Restormer.yml

Note: This training script uses 8 GPUs by default. To use any other number of GPUs, modify Restormer/train.sh and Denoising/Options/RealDenoising_Restormer.yml

Evaluation

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

Testing on SIDD dataset

  • Download SIDD validation data, run
python download_data.py --noise real --data test --dataset SIDD
  • To obtain denoised results, run
python test_real_denoising_sidd.py --save_images
  • To reproduce PSNR/SSIM scores on SIDD data (Table 6), run
evaluate_sidd.m

Testing on DND dataset

  • Download the DND benchmark data, run
python download_data.py --noise real --data test --dataset DND
  • To obtain denoised results, run
python test_real_denoising_dnd.py --save_images
  • To reproduce PSNR/SSIM scores (Table 6), upload the results to the DND benchmark website.