PyTorch implementation of “Light Field Image Super-Resolution with Transformers“, arXiv 2021. [pdf].


  • We make the first attempt to adapt Transformers to LF image processing, and propose a Transformer-based network for LF image SR.
  • We propose a novel paradigm (i.e., angular and spatial Transformers) to incorporate angular and spatial information in an LF.
  • With a small model size and low computational cost, our LFT achieves superior SR performance than other state-of-the-art methods.

Codes and Models:


  • PyTorch 1.3.0, torchvision 0.4.1. The code is tested with python=3.6, cuda=9.0.
  • Matlab (For training/test data generation and performance evaluation)


We used the EPFL, HCInew, HCIold, INRIA and STFgantry datasets for both training and test. Please first download our dataset via Baidu Drive (key:7nzy) or OneDrive, and place the 5 datasets to the folder ./datasets/.


  • Run Generate_Data_for_Training.m to generate training data. The generated data will be saved in ./data_for_train/ (SR_5x5_2x, SR_5x5_4x).
  • Run train.py to perform network training. Example for training LFT on 5×5 angular resolution for 4x/2xSR:

    $ python train.py --model_name LFT --angRes 5 --scale_factor 4 --batch_size 4
    $ python train.py --model_name LFT --angRes 5 --scale_factor 2 --batch_size 8
  • Checkpoint will be saved to ./log/.


  • Run Generate_Data_for_Test.m to generate test data. The generated data will be saved in ./data_for_test/ (SR_5x5_2x, SR_5x5_4x).
  • Run test.py to perform network inference. Example for test LFT on 5×5 angular resolution for 4x/2xSR:

    python test.py --model_name LFT --angRes 5 --scale_factor 4 \ 
    --use_pre_pth True --path_pre_pth './pth/LFT_5x5_4x_epoch_50_model.pth
    python test.py --model_name LFT --angRes 5 --scale_factor 2 \ 
    --use_pre_pth True --path_pre_pth './pth/LFT_5x5_2x_epoch_50_model.pth
  • The PSNR and SSIM values of each dataset will be saved to ./log/.


  • Quantitative Results

  • Efficiency

  • Visual Comparisons

  • Angular Consistency

  • Spatial-Aware Angular Modeling


If you find this work helpful, please consider citing:

    author    = {Liang, Zhengyu and Wang, Yingqian and Wang, Longguang and Yang, Jungang and Zhou, Shilin},
    title     = {Light Field Image Super-Resolution with Transformers},
    journal   = {arXiv preprint},
    month     = {August},
    year      = {2021},   


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