Skip to content

Self-Supervised 3D Scene Flow Estimation and Motion Prediction using Local Rigidity Prior (T-PAMI 2024). Motion prediction part.

License

Notifications You must be signed in to change notification settings

L1bra1/RigidFlowPP-Motion

Repository files navigation

RigidFlow++ (motion prediction part)

This is the PyTorch code for Self-Supervised 3D Scene Flow Estimation and Motion Prediction using Local Rigidity Prior (T-PAMI 2024). You can also check out the arXiv version at RigidFlowPP-arXiv.

In this repository, we apply RigidFlow++ for self-supervised motion prediction. For the codes in self-supervised 3D scene flow estimation, please refer to RigidFlowPP. The code is created by Ruibo Li ([email protected]).

Prerequisites

  • Python 3.7
  • NVIDIA GPU + CUDA CuDNN
  • PyTorch (torch == 1.9.0)

Create a conda environment for RigidPPMotion:

conda create -n RigidPPMotion python=3.7
conda activate RigidPPMotion
conda install pytorch==1.9.0 torchvision==0.10.0 cudatoolkit=10.2 -c pytorch
pip install numpy tqdm scikit-learn opencv-python matplotlib pyquaternion

Compile the furthest point sampling, grouping and gathering operation for PyTorch. We use the operation from this repo.

cd lib
python setup.py install
cd ../

Install & complie supervoxel segmentation method:

cd Supervoxel_utils
g++ -std=c++11 -fPIC -shared -o main.so main.cc
cd ../

More details about the supervoxel segmentation method, please refer to Supervoxel-for-3D-point-clouds.

Data preprocess

nuScenes

  1. Prepare the input data and the motion ground truth:

  2. Prepare raw point clouds and perform ground segmentation for self-supervision:

    • Extract raw point clouds and backward BEV maps for the training samples in nuScenes:
      python gen_data/gen_back_BEV_raw_point.py --root /path_to/nuScenes/nuScenes-data/ --split train --savepath /path_to/nuScenes/self-data/
      
    • Perform ground segmentation for raw point clouds:
      python gen_data/gen_ground_point.py --savepath /path_to/nuScenes/self-data/
      
      The data for self-supervised learning will be saved in /path_to/nuScenes/self-data/. Please references gen_data/README.md for more details.

Evaluation

Trained models

The trained model can be downloaded from model_nuScenes.pth.

Testing

Run the command:

python eval_SelfMotionNet.py --evaldata /path_to/nuScenes/input-data/test/ --pretrained pretrained/model_nuScenes.pth 

set evaldata to the directory of the test data (e.g., /path_to/nuScenes/input-data/test/).

set pretrained to the trained model (e.g., pretrained/model_nuScenes.pth).

Training

Run the command:

python train_SelfMotionNet.py --motiondata /path_to/nuScenes/input-data/train/ --selfdata /path_to/nuScenes/self-data/ --evaldata /path_to/nuScenes/input-data/val/ 

set motiondata to the directory of the input training data (e.g., /path_to/nuScenes/input-data/train/).

set selfdata to the directory of the self-supervised data (e.g., /path_to/nuScenes/self-data/).

set evaldata to the directory of the input validation or test data (e.g., /path_to/nuScenes/input-data/val/).

Citation

If you find this code useful, please cite our paper:

@article{li2024self,
  title={Self-Supervised 3D Scene Flow Estimation and Motion Prediction using Local Rigidity Prior},
  author={Li, Ruibo and Zhang, Chi and Wang, Zhe and Shen, Chunhua and Lin, Guosheng},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year={2024},
  publisher={IEEE}
}

Acknowledgement

Our project references the codes in the following repos.

About

Self-Supervised 3D Scene Flow Estimation and Motion Prediction using Local Rigidity Prior (T-PAMI 2024). Motion prediction part.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published