- GOOGLE HITNET model pytorch model will be released.
- GOOGLE HITNET model pytorch KITTI2015 submission: https://www.cvlibs.net/datasets/kitti/eval_scene_flow_detail.php?benchmark=stereo&result=226494ba5559e9f5f46bdbd681d1564fee78409e ranking 145 with 80GMAC
All the codes are tested in the following environment:
- Ubuntu 16.04
- Python 3.7
- PyTorch 1.1.0 or 1.2.0 or 1.3.0
- Torchvision 0.2.2 or 0.4.1
(1) Clone this repository.
git clone [email protected]:meteorshowers/X-StereoLab.git && cd X-StereoLab
(2) Setup Python environment.
conda activate -n xstereolab
pip install -r requirements.txt --user
## conda deactivate xstereolab
(1) Please download the KITTI dataset.
ln -s /path/to/KITTI_DATA_PATH ./data/kitti/
ln -s /path/to/OUTPUT_PATH ./outputs/
The training scripts support multi-processing distributed training, which is much faster than the typical PyTorch DataParallel interface.
python3 tools/train_net_disp.py --cfg ./configs/config_xxx.py --savemodel ./outputs/MODEL_NAME -btrain 4 -d 0-3 --multiprocessing-distributed
The training models, configuration and logs will be saved in the model folder.
To load some pretrained model, you can run
python3 tools/train_net_disp.py --cfg xxx/config.py --loadmodel ./outputs/MODEL_NAMEx --start_epoch xxx --savemodel ./outputs/MODEL_NAME -btrain 4 -d 0-3 --multiprocessing-distributed
If you want to continue training from some epochs, just set the cfg, loadmodel and start_epoch to the respective model path.
Besides, you can start a tensorboard session by
tensorboard --logdir=./outputs/MODEL_NAME/tensorboard --port=6666
and visualize your training process by accessing https://localhost:6666 on your browser.
on working ...
</tbody>
Methods | Epochs | Train Mem (GB/Img) | Test Mem (GB/Img) | EPE | D1-all | Models |
---|---|---|---|---|---|---|
HITNET (kitti) | 4200 | 2.43% | GoogleDrive | |||
HITNET (sceneflow) | 200 | 0.65 | GoogleDrive | |||
stereonet (sceneflow) | 20 | 1.10 | GoogleDrive | |||
ActiveStereoNet | 10 | GoogleDrive | ||||
SOS |
Methods | Epochs | Train Mem (GB/Img) | Test Mem (GB/Img) | 3D BEV AP (Ours small plume) | 3D BEV AP (Paper small plume) |
---|---|---|---|---|---|
PLUME | 72.9 / 62.5 / 56.9 | 74.4 / 61.7 / 55.8 |
We provide a video demo for showing the result of X-StereoLab. Here we show the predicted disparity map of activastereonet.
- Multiprocessing GPU training
- TensorboardX
- Reduce training GPU memory usage
- eval and test code
- Result visualization
- Still in progress
If you find our work useful in your research, please consider citing:
@misc{XStereoLab2021,
title={{X-StereoLab} stereo matching and stereo 3D object detection toolbox},
author={X-StereoLab Contributors},
howpublished = {\url{https://github.com/meteorshowers/X-StereoLab}},
year={2021}
}
* refercence[2]
@article{tankovich2020hitnet,
title={HITNet: Hierarchical Iterative Tile Refinement Network for Real-time Stereo Matching},
author={Tankovich, Vladimir and H{\"a}ne, Christian and Fanello, Sean and Zhang, Yinda and Izadi, Shahram and Bouaziz, Sofien},
journal={arXiv preprint arXiv:2007.12140},
year={2020}
}
* refercence[3]
@inproceedings{tankovich2018sos,
title={Sos: Stereo matching in o (1) with slanted support windows},
author={Tankovich, Vladimir and Schoenberg, Michael and Fanello, Sean Ryan and Kowdle, Adarsh and Rhemann, Christoph and Dzitsiuk, Maksym and Schmidt, Mirko and Valentin, Julien and Izadi, Shahram},
booktitle={2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
pages={6782--6789},
year={2018},
organization={IEEE}
}
vtankovich |
Yan Wang Waymo |
- Thanks to SamehKhamis (NVIDIA)
The code is released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International Public License for NonCommercial use only. Any commercial use should get formal permission first.
If you have any questions or suggestions about this repo, please feel free to contact me ([email protected]). Wechat:
![]() XUANYILI |