Group-wise Correlation Stereo Network, CVPR 2019

Go to file

Xiaoyang Guo 0987074f48 add scene flow pretrained model link		2019-08-08 22:51:23 -07:00
datasets	add codes for saving submission images	2019-04-14 21:51:10 +08:00
filenames	init commit	2019-04-14 17:34:58 +08:00
models	add math package back	2019-04-14 09:28:06 -04:00
scripts	move scripts into a new folder	2019-04-15 23:12:26 +08:00
utils	remove unused packages	2019-04-14 21:26:51 +08:00
.gitignore	init commit	2019-04-14 17:34:58 +08:00
LICENSE	add LICENSE	2019-04-17 19:24:38 +08:00
main.py	fix import bug	2019-04-14 09:29:58 -04:00
README.md	add scene flow pretrained model link	2019-08-08 22:51:23 -07:00
save_disp.py	add codes for saving submission images	2019-04-14 21:51:10 +08:00

README.md

GwcNet

This is the implementation of the paper Group-wise Correlation Stereo Network, CVPR 19, Xiaoyang Guo, Kai Yang, Wukui Yang, Xiaogang Wang, and Hongsheng Li [Arxiv]

How to use

Environment

python 3.6
Pytorch >= 0.4.1

Data Preparation

Download Scene Flow Datasets, KITTI 2012, KITTI 2015

Training

Scene Flow Datasets

run the script ./scripts/sceneflow.sh to train on Scene Flow datsets. Please update DATAPATH in the bash file as your training data path.

KITTI 2012 / 2015

run the script ./scripts/kitti12.sh and ./scripts/kitti15.sh to finetune on the KITTI 12/15 dataset. Please update DATAPATH and --loadckpt as your training data path and pretrained SceneFlow checkpoint file.

Evaluation

run the script ./scripts/kitti12_save.sh and ./scripts/kitti15_save.sh to save png predictions on the test set of the KITTI datasets to the folder ./predictions.

Pretrained Models

Scene Flow KITTI 2012/2015

Citation

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

@inproceedings{guo2019group,
  title={Group-wise Correlation Stereo Network},
  author={Guo, Xiaoyang and Yang, Kai and Yang, Wukui and Wang, Xiaogang and Li, Hongsheng},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  pages={3273--3282},
  year={2019}
}

Acknowledgements

Thanks to Jia-Ren Chang for opening source of his excellent work PSMNet. Our work is inspired by this work and part of codes in models are migrated from PSMNet.