roj4s/pan

[Params: Only 272K!!!] Efficient Image Super-Resolution Using Pixel Attention, in ECCV Workshop, 2020.

If you have questions about results, please check the new update version of file PAN_arch.py.

PAN [:zap: 272K parameters]

Lowest parameters in AIM2020 Efficient Super Resolution.

Paper | Video

Efficient Image Super-Resolution Using Pixel Attention

Authors: Hengyuan Zhao, Xiangtao Kong, Jingwen He, Yu Qiao, Chao Dong

Dependencies

• Python >= 3.6 (Recommend to use Anaconda)
• PyTorch >= 1.5.0
• NVIDIA GPU + CUDA
• Python packages: pip install numpy opencv-python lmdb
• [option] Python packages: pip install tensorboardX, for visualizing curves.

Codes

• Our codes version based on mmsr.
• This codes provide the testing and training code.

How to Test

1. Clone this github repo.

git clone https://github.com/zhaohengyuan1/PAN.git
cd PAN

2. Download the five test datasets (Set5, Set14, B100, Urban100, Manga109) from Google Drive

3. Pretrained models have be placed in ./experiments/pretrained_models/ folder. More models can be download from Google Drive.

4. Run test. We provide x2,x3,x4 pretrained models.

cd codes
python test.py -opt option/test/test_PANx4.yml

More testing commond can be found in ./codes/run_scripts.sh file. 5. The output results will be sorted in ./results. (We have been put our testing log file in ./results) We also provide our testing results on five benchmark datasets on Google Drive or Baidu Drive, password: 8mrn.

How to Train

1. Download DIV2K and Flickr2K from Google Drive or Baidu Drive

2. Generate Training patches. Modified the path of your training datasets in ./codes/data_scripts/extract_subimages.py file.

3. Run Training.

python train.py -opt options/train/train_PANx4.yml

4. More training commond can be found in ./codes/run_scripts.sh file.

Testing the Parameters, Mult-Adds and Running Time

1. Testing the parameters and Mult-Adds.

python test_summary.py

2. Testing the Running Time.

python test_running_time.py

Related Work on AIM2020

Enhanced Quadratic Video Interpolation (winning solution of AIM2020 VTSR Challenge) paper | code

Contact

Email: [email protected]