WDSR
This implements training of WDSR on the DIV2K_x2 dataset.
- Reference impkementation:
url=https://github.com/ychfan/wdsr
branch=master
commit_id=b78256293c435ef34e8eab3098484777c0ca0e10
WDSR Detail
For Details, see src/models/wdsr.py
Requirements
- Install PyTorch (pytorch.org)
pip install -r requirements.txtNote: pillow recommends installing a newer version. If the corresponding torchvision version cannot be installed directly, you can use the source code to install the corresponding version. The source code reference link: Suggestion the pillow is 9.1.0 and the torchvision is 0.6.0- The DIV2k Dataset can be downloaded from Reference impkementation (readme), find
DIV2K dataset: DIVerse 2K resolution high quality images as used for the NTIRE challenge on super-resolution @ CVPR 2017link , downloadTrain Data (HR images),Validation Data (HR images),Train Data Track 1 bicubic downscaling x2 (LR images),Validation Data Track 1 bicubic downscaling x2 (LR images).Move the datasets to directory ./data/DIV2K/
Training
To train a model, run trainer.py with the desired model architecture and the path to the DIV2K dataset:
# 1p training full
# 备注: 目标精度34.75;验收精度35.3625
bash test/train_full_1p.sh
# 1p train perf
bash test/train_performance_1p.sh
# 1p testing
bash test/eval_1p.sh
# 8p training full
# 备注: 目标精度34.75;验收精度33.7371
bash test/train_full_8p.sh
# 8p train perf
bash test/train_performance_8p.sh
# 8p testing
bash test/train_eval_8p.sh --pre_train_model=real_pre_train_model
# demo
# 请将要测试图片路径作为lr_image的参数传入,输出会在output_sr 文件夹
python3 demo.py --pre_train_model real_pre_train_model --lr_image 0801x2.png
wdsr training result
| PSNR (dB) | Npu_nums | Epochs | AMP_Type |
|---|---|---|---|
| 35.3625 | 1 | 30 | O2 |
| 34.3368 | 8 | 30 | O2 |