lancelet / tinyhitnet Goto Github PK

This is a Pytorch implementations of "HITNet: Hierarchical Iterative Tile Refinement Network for Real-time Stereo Matching".

python3 -m venv venv
source venv/bin/activate
pip install torch

1. Compile and install cuda op

   CC=gcc-10 CXX=g++-10 pip install ./ext_op

2. Replace dataset path in preprocess/plane_fitting.py and script/hitnet_sf_finalpass.sh

3. Robust plane fitting

   python -m preprocess.plane_fitting_sf
   

4. Training

   bash script/hitnet_sf_finalpass.sh

1. Replace dataset path in eval.py

2. Evaluation

   python eval.py --model HITNet --ckpt ckpt/{ckpt_name} --data_type SceneFlow --data_root_val {path} --data_list_val lists/sceneflow_test.list

python predict.py --model HITNet --ckpt ckpt/{ckpt_name} --images {left.png} {right.png} --output {disp.png}

@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}
}

However, if you find this implementation or pre-trained models helpful, please consider to cite:

@misc{hang2021tinyhitnet,
  title={TinyHITNet},
  author={zjjMaiMai},
  howpublished={\url{https://github.com/zjjMaiMai/TinyHITNet}},
  year={2021}
}