Awesome
LKDN
Large Kernel Distillation Network for Efficient Single Image Super-Resolution
Chengxing Xie, Xiaoming Zhang, Linze Li, Haiteng Meng, Tianlin Zhang, Tianrui Li and Xiaole Zhao
Environment
- PyTorch >= 1.7 (Recommend >= 1.11)
- BasicSR = 1.4.2
Installation
pip install -r requirements.txt
python setup.py develop
How To Test
- Refer to
./options/test/LKDNfor the configuration file of the model to be tested, and prepare the testing data and pretrained model. - The pretrained models are available in
./experiments/pretrained_models/LKDN. - Then run the follwing codes (taking
LKDN_x4.pthas an example):
python basicsr/test.py -opt options/test/LKDN/test_LKDN_x4.yml
The testing results will be saved in the ./results folder.
- Refer to
./inferencefor inference without the ground truth image. - Refer to
./basicsr/calculate_params_flops.pyfor calculating the parameters and flops.
How To Train
- Refer to
./options/trainfor the configuration file of the model to train. - Preparation of training data can refer to this page.
- The training command is like:
python basicsr/train.py -opt options/train/LKDN/train_LKDN_x4.yml
More training commands can refer to this page.
The training logs and weights will be saved in the ./experiments folder.
How To Re-parameterize
Refer to ./pth for the validation and use of re-parameterization.
conv1x1_3x3.py, conv1x1.py and shortcut.py respectively verify the three re-parameterization methods.
del_params_ema.py simplifies the .pth file. (Remove the additional parameters retained when using EMA.)
print_pth.py prints the content of the .pth file.
reparm.py reparameterizes the model.
Note that the LKDN-S_del_rep_x4.pth is the model after re-parameterizing, and the LKDN-S_x4.pth is the model without re-parameterizing.
Results
Benchmark results on SR ×4. Multi-Adds is calculated with a 1280 × 720 GT image.
| Method | Params[K] | Multi-Adds[G] | Set5 PSNR/SSIM | Set14 PSNR/SSIM | BSD100 PSNR/SSIM | Urban100 PSNR/SSIM | Manga109 PNSR/SSIM |
|---|---|---|---|---|---|---|---|
| BSRN | 352 | 19.4 | 32.35/0.8966 | 28.73/0.7847 | 27.65/0.7387 | 26.27/0.7908 | 30.84/0.9123 |
| VapSR | 342 | 19.5 | 32.38/0.8978 | 28.77/0.7852 | 27.68/0.7398 | 26.35/0.7941 | 30.89/0.9132 |
| LKDN | 322 | 18.3 | 32.39/0.8979 | 28.79/0.7859 | 27.69/0.7402 | 26.42/0.7965 | 30.97/0.9140 |
The inference results on benchmark datasets are available at Google Drive or Baidu Netdisk.
Contact
If you have any question, please email zxc0074869@gmail.com.