ATTSF Attention! Stay Focus!

Solution of Defocus Deblurring Challenge

Attention! Stay Focus! (ATTSF) - NTIRE 2021 - Paper

by Tu Vo

Content

• ATTSF Attention! Stay Focus!
• Getting Started
• Running
• References
• Citations

Getting Started

• Clone the repository

Prerequisites

• Tensorflow 2.2.0+
• Tensorflow_addons
• Python 3.6+
• Keras 2.3.0
• PIL
• numpy

Running

Training

• Preprocess

  • Download the training data

  • Unzip the file

• Train ATTSF

  • change op_phase='train' in config.py

  python main.py
  

• Test ATTSF

  • change op_phase='valid' in config.py

  python main.py
  

Usage

Training

usage: main.py [-h] [--filter FILTER] [--attention_filter ATTENTION_FILTER]
[--kernel KERNEL] [--encoder_kernel ENCODER_KERNEL]
[--decoder_kernel DECODER_KERNEL]
[--triple_pass_filter TRIPLE_PASS_FILTER] [--num_rrg NUM_RRG]
[--num_mrb NUM_MRB]

optional arguments:
-h, --help show this help message and exit
--filter FILTER
--attention_filter ATTENTION_FILTER
--kernel KERNEL
--encoder_kernel ENCODER_KERNEL
--decoder_kernel DECODER_KERNEL
--triple_pass_filter TRIPLE_PASS_FILTER

Testing

• Download the weight here and put it to the folder ModelCheckpoints
• Note: as the part of our research, the weight file has been hidden.

Result

 Left image | Right Image | Output

License

This project is licensed under the MIT License - see the LICENSE file for details

References

[1] Defocus Deblurring Challenge - NTIRE2021

Citation

@InProceedings{Vo_2021_CVPR,
author = {Vo, Tu},
title = {Attention! Stay Focus!},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2021},
pages = {479-486}
}

Acknowledgments

• This work is heavily based on the code from the challenge host . Thank you for the hard job.