A deep recursive multi-scale feature fusion network for image super-resolution

This repository is Pytorch code for our proposed DRMSFFN. The structure of the proposed Deep Recursive Multi-Scale Feature Fusion Network (DRMSFFN). Structure of the Recursive Multi-Scale Feature Fusion Block (RMSFFB).

The details about our proposed DRMSFFN can be found in our paper:https://www.sciencedirect.com/science/article/pii/S1047320322002504

If you find our work useful in your research work, please star the code and consider citing:

@article{liu2023deep,
  title={A deep recursive multi-scale feature fusion network for image super-resolution},
  author={Liu, Feiqiang and Yang, Xiaomin and De Baets, Bernard},
  journal={Journal of Visual Communication and Image Representation},
  volume={90},
  pages={103730},
  year={2023},
  publisher={Elsevier}
}

Requirements:

1. Python==3.6 (Anaconda is recommended)
2. skimage
3. imageio
4. Pytorch==1.2
5. tqdm
6. pandas
7. cv2 (pip install opencv-python)

Test:

python test.py -opt options/test/test_DRMSFFN_x2.json
python test.py -opt options/test/test_DRMSFFN_x3.json
python test.py -opt options/test/test_DRMSFFN_x4.json

Finally, PSNR/SSIM values for Set5 are shown on your screen, you can find the reconstruction images in ./results. Other standard SR benchmark dadasets, you need to
change the datasets storage path in the test_DRMSFFN_x2.json, test_DRMSFFN_x3.json and test_DRMSFFN_x4.json files.

Results:

Quantitative Results: Quantitative comparison for scale factors ×2, ×3 and ×4 of the proposed method DRMSFFN with state-of-the-art methods. The best and the second best results are indicated in bold and underlined, respectively.

Some Qualitative Results: Visual comparison of the results of the proposed method DRMSFFN with those of other state-of-the-art methods on some images from the Set14 and B100 datasets for ×4 SR. The best results are indicated in bold.

Visual comparison of the results of our DRMSFFN with those of other state-of-the-art methods on some images from the Urban100 and Manga109 datasets for ×4 SR. The best results are indicated in bold.

Model Analysis The running time, number of parameters and performance of different SISR methods. The results denote the mean PSNR and running time for scale factor x4 on the Set5 dataset.