Skip to content

huzexi/DKNet

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

DKNet

This repo is the implementation of Decomposition Kernel Network (DKNet) for light field super-resolution.

Texture-Enhanced Light Field Super-Resolution With Spatio-Angular Decomposition Kernels
Zexi Hu; Xiaoming Chen; Henry Wing Fung Yeung; Yuk Ying Chung; Zhibo Chen
IEEE Transactions on Instrumentation and Measurement (2022)
[IEEE Xplore] [arXiv]

Citation

Please cite us if it is helpful for your work:

@article{hu2022texture,
  title={Texture-Enhanced Light Field Super-Resolution With Spatio-Angular Decomposition Kernels},
  author={Hu, Zexi and Chen, Xiaoming and Yeung, Henry Wing Fung and Chung, Yuk Ying and Chen, Zhibo},
  journal={IEEE Transactions on Instrumentation and Measurement},
  volume={71},
  pages={1--16},
  year={2022},
  publisher={IEEE}
}

How to run

Dependencies

Most dependencies have been listed in requirements.txt, simply run the following command to install them:

pip install -r requirements.txt

The code is tested with Python 3.8/3.10 and Ubuntu 20.04/22.04.

Data preparation

This step is to loading the huge LF images into HDF5 files, which simplifies the process and accelerate the speed when training.

Refer to the following webpages to download the datasets firstly:

  1. SIGASIA16: https://cseweb.ucsd.edu/~viscomp/projects/LF/papers/SIGASIA16/
  2. Stanford: http://lightfields.stanford.edu/LF2016.html

After then, modify the paths in component/dataset/LytroDataset.py to the corresponding locations, e.g. Stanford and SIGASIA16, and run the following command:

python prepare_dataset.py

The command uses component/dataset/YeungDataset.py, which inherits from LytroDataset and specify samples for the training and testing set as in Yeung's work.

Please be reminded that this step will read the LF images into HDF5 files which might be big. The Yeung dataset is nearly 40GB.

Testing

Run the following command will infer the given model on the prepared dataset calculating PSNR and SSIM and producing images in workdir/test directory.

python test.py --model /path/to/weight.pt --bgr

Refer to the Release page for the pretrained weight files.

Training

Run the following command to launch a training session named train1.

python train.py --name train1

A folder will be created in workdir/train containing all the files generated during the training, including weight files, logs, configuration, etc. You can also run the following command to monitor the training process using tensorboard.

tensorboard --logdir workdir/train

Configuration

Readers are suggested to check the following configuration files for customisation.

  • component/config/DefaultConfig.py includes most of the configuration variables as default, component/config/CustomConfig.py overrides the default and is used when running.
  • component/network/DKNet/DKNetConfig.py includes the configuration of DKNet.

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages