Code for the paper:
G. Wang, X. Tian, R. Ding, and H. Wang, Unsupervised Learning of Scene Flow from Monocular Camera, in International Conference on Robotics and Automation, 2021.
Our model is trained and tested under:
- Python 3.6.9
- NVIDIA GPU + CUDA CuDNN
- PyTorch (torch == 1.7.1)
- torchvision
- scipy
- argparse
- tensorboardX
- tqdm
- numba
- cffi
Operation in this repo is used to compile the furthest point sampling, grouping and gathering operation for PyTorch.
cd pointnet2
python setup.py install
cd ../We referred to the dataset pre-processing approach in this repo to generate our data for training from KITTI Odometry dataset. We uploaded our processed dataset and you can click this link to download.
KITTI scene flow dataset is used for test and evaluation. Flownet3d processed the first 150 data points from KITTI scene flow dataset and removed the ground points. We copied the link here for you to download.
Set data in the configuration file to the path where your training data is saved and change the batch_size according to your GPU. Then run:
python train.py config.yamlOur pre-trained models are provided here for download.
Set evaluate in the configuration file to be True. Then run:
python train.py config.yamlWe are grateful for repo for its github repository.