Pytorch implementation of "Exploring Sparsity in Image Super-Resolution for Efficient Inference", CVPR 2021

[arXiv]

• Locate and skip redundant computation in SR networks at a fine-grained level for efficient inference.
• Maintain state-of-the-art performance with significant FLOPs reduction and a speedup on mobile devices.
• Efficient implementation of sparse convolution based on original Pytorch APIs for easier migration and deployment.

For easier migration and deployment, we use an efficient implementation of sparse convolution based on original Pytorch APIs rather than the commonly applied CUDA-based implementation. Specifically, sparse features are first extracted from the input, as shown in the following figure. Then, matrix multiplication is executed to produce the output features.

• Python 3.6
• PyTorch == 1.1.0
• numpy
• skimage
• imageio
• matplotlib
• cv2

1.1 Download DIV2K training data (800 training + 100 validtion images) from DIV2K dataset or SNU_CVLab.

1.2 Specify '--dir_data' based on the HR and LR images path. In option.py, '--ext' is set as 'sep_reset', which first convert .png to .npy. If all the training images (.png) are converted to .npy files, then set '--ext sep' to skip converting files.

For more informaiton, please refer to EDSR(PyTorch).

python main.py --model SMSR --save SMSR_X2 --scale 2 --patch_size 96 --batch_size 16

Download benchmark datasets (e.g., Set5, Set14 and other test sets) and prepare HR/LR images in testsets/benchmark following the example of testsets/benchmark/Set5.

python main.py --dir_data testsets --data_test Set5 --scale 2 --model SMSR --save SMSR_X2 --pre_train experiment/SMSR_X2/model/model_1000.pt --test_only --save_results

@InProceedings{Wang2020Exploring,
  author    = {Wang, Longguang and Dong, Xiaoyu and Wang, Yingqian and Ying, Xinyi and Lin, Zaiping and An, Wei and Guo, Yulan},
  title     = {Exploring Sparsity in Image Super-Resolution for Efficient Inference},
  booktitle = {CVPR},
  year      = {2021},
}

This code is built on EDSR (PyTorch). We thank the authors for sharing the codes.