Awesome

RR-HLFSR_NTIRE2023_LFSR

Residual in Resiual Learning based Hybrid Light Field Image Super-Resolution Network (RR-HLFSR)

News

[2023-03-20]: We released our RR-HLFSR, which is participated in NTIRE2023. Our RR-HLFSR is an enhanced version of our HLFSR

Results

We share the pre-trained models and the SR LF images generated by our RR-HLFSR model for 4x LF spatial SR, which are avaliable at https://drive.google.com/drive/u/2/folders/160KS4l5jWEehJ0KtgOg0T6pdlhgTyWbg

Code

Dependencies

Python 3.6
Pyorch 1.3.1 + torchvision 0.4.2 + cuda 92
Matlab

This code is built up on the Basic-LFSR. Please go to orginal code for more guideline information.

Dataset

We use the processed data by LF-DFnet, including EPFL, HCInew, HCIold, INRIA and STFgantry datasets for training and testing. Please download the dataset in the official repository of LF-DFnet.

Prepare Training and Test Data

To generate the training data, please first download the five datasets and run:
```
Generate_Data_for_Training.m
```
To generate the test data, run:
```
Generate_Data_for_Test.m
```

Train

Run:

python train.py  --model_name RR_HLFSR --angRes 5 --scale_factor 4 --n_groups 10 --n_blocks 15 --channels 64  --crop_test_method 3

Test

Run:

python test.py --model_name RR_HLFSR --angRes 5 --upscale_factor 4 --n_groups 10 --n_blocks 15 --channels 64  --crop_test_method 1 --self_ensemble True  --model_path [pre-trained dir]

[Important note]:

For our HLFSR method, the performance is following “the larger image patch size is the better”. For example, if we keep the whole image as an input of our network (i.e., crop_test_method is fixed equal to 1), it can be achieved the best performance. This is because our proposed network components require an adequate size of an input image to better exploit the pixel correlations in a larger receptive field. To get the same performance as reported in this NTIRE2023 LFSR challenge, we need to set the default crop_test_method equal to 1.
We use the geometric self-ensemble method to further improve the performance in NTIRE2023 LFSR challenges
We may need to turn off the calculate PSNR/SSIM by settings "--test_NTIRE2023_LFSR 1" since there is no ground true HR images during the testing phase.

Citation

If you find this work helpful, please consider citing the following papers:

@Article{vinh2023-lfsr,
  author  = {Duong, V. V. and Nguyen, T. H. and Yim, J. and Jeon, B.},
  journal = {IEEE Trans. Compuational Imaging},
  title   = {Light Field Image Super-Resolution Network via Joint Spatial-Angular and Epipolar Information},
  year    = {2023},
}

@InProceedings{NTIRE2023-LFSR,
  author    = {Wang, Yingqian and Wang, Longguang and Liang, Zhengyu and Yang, Jungang and Timofte, Radu and Guo, Yulan},
  title     = {NTIRE 2023 Challenge on Light Field Image Super-Resolution},
  booktitle = {CVPRW},
  year      = {2023},
}

Acknowledgement

Our work and implementations are inspired and based on the following projects: Basic-LFSR LF-DFnet LF-InterNet We sincerely thank the authors for sharing their code and amazing research work!

Contact

if you have any question, please contact me through email duongvinh@skku.edu