Official pytorch implementation of paper Single image super-resolution based on trainable feature matching attention network (TFMAN) (Google Scholar).

The organization structure of the code is very simple. You can directly run python test.py to test the SR performance of TFMAN model trained with x2 Bicubic degradation model on Set5 and Set14 datasets.

Please unzip the downloaded training dataset (DIV2K / DIV2K_BD / DIV2K_DN, see below) into the directory SRTrain. Then, you can train TFMAN by just running: python train.py

You can download the TFMAN checkpoints and the training/testing datasets from the following website:

If you want to train or test custom dataset, please organize the directory structure of the dataset into a format consistent with our datasets and run train.py or test.py.

If this paper is fortunate enough to help improve your work, please cite us. [Code][Paper]

@article{TFMAN,
    title = {Single image super-resolution based on trainable feature matching attention network},
    journal = {Pattern Recognition},
    year = {2024},
    doi = {https://doi.org/10.1016/j.patcog.2024.110289},
    url = {https://www.sciencedirect.com/science/article/pii/S0031320324000402},
    author = {Qizhou Chen and Qing Shao},
}