Home

Awesome

ITSA (CVPR 2022)

This is the official code of the work ITSA: An Information-Theoretic Approach to Automatic Shortcut Avoidance and Domain Generalization in Stereo Matching Networks, CVPR 2022,
WeiQin Chuah, Ruwan Tennakoon, Reza Hoseinnezhad, Ali Bab-Hadiashar, David Suter. [Arxiv]

How to use:

Environment setup

:package: Dataset

:clock4: Training

Run bash script/train.sh for training.

Arguments

:memo: Inference

Run bash eval_kitti.sh for inference. Predicted disparity maps for the chosen dataset will be saved to the directory set in the argument savepath.

Arguments

Example

To run inference using KITTI 2015 dataset and PSMNet, run the following command:

python infer.py --model PSMNet \
                --savepath {path_to_save_images} \
                --loadmodel {path_to_pretrained_weights}\
                --kitti15 {change this to the desired dataset}

Make sure the directory to the datasets are correct.

Pretrained Weights [Google Drive]

Citation

If you find this code useful in your research, please cite:

@InProceedings{Chuah_2022_CVPR,
    author    = {Chuah, WeiQin and Tennakoon, Ruwan and Hoseinnezhad, Reza and Bab-Hadiashar, Alireza and Suter, David},
    title     = {ITSA: An Information-Theoretic Approach to Automatic Shortcut Avoidance and Domain Generalization in Stereo Matching Networks},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2022},
    pages     = {13022-13032}
}

Acknowledgement

The codes in this work heavily relies on codes by PSMNet, GwcNet and CFNet. We thank the original authors for their awesome repos.