Stripformer: Strip Transformer for Fast Image Deblurring (ECCV 2022 Oral)

Pytorch Implementation of "Stripformer: Strip Transformer for Fast Image Deblurring"

Installation

The implementation of our BANet is modified from "DeblurGANv2"

git clone https://github.com/pp00704831/Stripformer.git
cd Stripformer
conda create -n Stripformer python=3.6
source activate Stripformer
conda install pytorch==1.8.0 torchvision==0.9.0 torchaudio==0.8.0 cudatoolkit=11.1 -c pytorch -c conda-forge
pip install opencv-python tqdm pyyaml joblib glog scikit-image tensorboardX albumentations==1.1.0
pip install -U albumentations[imgaug]

Training

Download "GoPro" dataset into './datasets'
For example: './datasets/GoPro/train/blur/**/*.png'

We train our Stripformer in two stages:
1) We pre-train Stripformer for 3000 epochs on patch size 256x256. Please run the following commands.

python pretrained.py

2) After stage 1, we keep training Stripformer for 1000 epochs on patch size 512x512. Please run the following commands.

python train.py

Testing

For reproducing our results on GoPro and HIDE dataset, download the "Stripformer_gopro.pth"

For reproducing our results on RealBlur dataset, download "Stripformer_realblur_J.pth" and "Stripformer_realblur_R.pth"

• For testing on GoPro test set
  Download "GoPro" full dataset or test set into './datasets'
  For example: './datasets/GoPro/test/blur/**/*.png'

python predict_GoPro_test_results.py --weights_path ./Stripformer_gopro.pth 

• For testing on HIDE dataset
  Download "HIDE" into './datasets'
  

python predict_HIDE_results.py --weights_path ./Stripformer_gopro.pth 

• For testing on RealBlur test sets Download "RealBlur_J" and "RealBlur_R" into './datasets'
  

python predict_RealBlur_J_test_results.py --weights_path ./Stripformer_realblur_J.pth 

python predict_RealBlur_R_test_results.py --weights_path ./Stripformer_realblur_R.pth 

• For testing your own training weight (take GoPro for a example)

1. Rename the path in line 23 in the predict_GoPro_test_results.py
   
2. Chage command to --weights_path ./final_Stripformer_gopro.pth

Evaluation

• For evaluation on GoPro results in MATLAB, download "Stripformer_GoPro_results" into './out'

evaluation_GoPro.m

• For evaluation on HIDE results in MATLAB, download "Stripformer_HIDE_results" into './out'

evaluation_HIDE.m

• For evaluation on RealBlur_J results, download "Stripformer_realblur_J_results" into './out'

python evaluate_RealBlur_J.py

• For evaluation on RealBlur_R results, download "Stripformer_realblur_R_results" into './out'

python evaluate_RealBlur_R.py