Official PyTorch code for the TBC2024 paper "MAFBLiF: Multi-scale Attention Feature Fusion Based Blind Light Field Image Quality Assessment". Please refer to the paper for details.

Note: First, we convert the dataset into H5 files using MATLAB. Then, we train and test the model in Python.

Generate Dataset in MATLAB

Take the NBU-LF1.0 dataset for instance, convert the dataset into h5 files, and then put them into './Datasets/NBU_MLI_7x32x32/':

 ./MAFBLiF/Datasets/Generateh5_for_NBU_Dataset.m

Train

Train the model using the following command:

python Train.py  --trainset_dir ./Datasets/NBU_MLI_7x32x32/

Test Overall Performance

Test the overall performance using the following command:

python Test.py

Test Individual Distortion Type Performance

Test the individual distortion type performance using the following command:

 python Test_Dist.py

Acknowledgement

This project is based on DeeBLiF. Thanks for the awesome work.

Citation

Please cite the following paper if you use this repository in your reseach.

@ARTICLE{10623345,
  author={Zhou, Rui and Jiang, Gangyi and Cui, Yueli and Chen, Yeyao and Xu, Haiyong and Luo, Ting and Yu, Mei},
  journal={IEEE Transactions on Broadcasting}, 
  title={MAFBLiF: Multi-Scale Attention Feature Fusion-Based Blind Light Field Image Quality Assessment}, 
  year={2024},
  volume={},
  number={},
  pages={1-13},
  keywords={Measurement;Feature extraction;Image quality;Visualization;Tensors;Electronic mail;Distortion measurement;Light field;blind image quality assessment;multi-scale attention;spatial-angular features;pooling},
  doi={10.1109/TBC.2024.3434699}}

Contact

For any questions, feel free to contact: [email protected] or [email protected]