LF-InterNet

PyTorch implementation of "Spatial-Angular Interaction for Light Field Image Super-Resolution", ECCV 2020.

News: We recommend our newly-released repository BasicLFSR for the implementation of our LF-InterNet. BasicLFSR is an open-source and easy-to-use toolbox for LF image SR. A number of milestone methods have been implemented (retrained) in a unified framework in BasicLFSR.

Network Architecture:

Codes and Models:

Requirement:

• PyTorch 1.3.0, torchvision 0.4.1. The code is tested with python=3.7, cuda=9.0.
• Matlab (For training/test data generation and performance evaluation)

Test:

• Download the test sets and unzip them to ./data. Here, we provide a demo test set (data_demo.zip) which only includes one test scene, and we also provide the full test set on Baidu Drive (Key: NUDT) which is used in our paper.
• Download our pretrained models (log.zip) and unzip them to ./log.
• Run GenerateDataForTest.m to generate test data.
• Run test.py to perform a demo inference. Note that, the selected pretrained model should match the generated input data and the preset network architecture. Initial results (.mat files) will be saved to ./results.
• Run evaluation.m to calculate PSNR and SSIM scores and transform initial results (.mat files) into .png images.

Train:

Please switch to LF-InterNet_train for details.

Results:

Quantitative Results:

Visual Comparisons:

Efficiency:

Performance w.r.t. Perspectives:

Performance on Unseen Datasets:

Performance on Real LFs:

Citiation

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

@InProceedings{LF-InterNet,
  author    = {Wang, Yingqian and Wang, Longguang and Yang, Jungang and An, Wei and Yu, Jingyi and Guo, Yulan},
  title     = {Spatial-Angular Interaction for Light Field Image Super-Resolution},
  booktitle = {European Conference on Computer Vision (ECCV)},
  pages     = {290-308},
  year      = {2020},
}

Contact

Any question regarding this work can be addressed to [email protected].