ACLRNet

PyTorch implementation of the paper Stereo Image Restoration via Attention-Guided Correspondence Learning

Description

ACLRNet is a stereo image restoration framework, including stereo image denoising, super-resolution and compression artifact reduction.

Requirement:

• Python 3.7.9, skimage 0.16.2, PyTorch 1.8.1, torchvision 0.9.1 and CUDA 10.2
• Matlab (For training data generation)

Train:

• Download the training dataset Flickr1024_train from Baidu Drive (Key: zk3m) and unzip ititit to ./data/train.
• Run ./data/GenerateTrainingPatches.m to generate training patches. The scales are set to 1, 2, 4 for different restoration tasks.
• Run train.py to perform training. Checkpoint will be saved to ./log/.

Test:

• Download the testing datasets (KITTI2012, KITTI2015, Middlebury, ETH3D and Flickr1024_test) from Baidu Drive (Key: zm1p) and unzip them to ./data/test.
• Run test.py to perform inference. Results (.png files) will be saved to ./results.

Citiation:

If you find our work useful for your research, please consider citing this paper:

@article{zhang2024stereo,
  title={Stereo Image Restoration Via Attention-Guided Correspondence Learning},
  author={Zhang, Shengping and Yu, Wei and Jiang, Feng and Nie, Liqiang and Yao, Hongxun and Huang, Qingming and Tao, Dacheng},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year={2024},
  publisher={IEEE}
}

Acknowledgement:

Our code is inspired by IPASSR. We thank the authors for their great job! For questions, please send an email to [email protected].