TorchBox3D

torchbox3d is a 3D perception library for autonomous driving datasets.

Overview

• Currently supports 3D object detection on the Argoverse 2 Sensor dataset.
• Native pytorch-lightning support.
• Multi-gpu training.

Supported Models

• Architectures:

  • SECOND [Sensors 2018]

• Heads:

  • PointPillars [CVPR 2019]
  • CenterPoint [CVPR 2021]

Supported Datasets

• Argoverse 2 Sensor Dataset [NeurIPS Datasets and Benchmarks]

Installation

---

Source Install

This will install torchbox3d as a conda package.

bash conda/install.sh

Configuration

---

Configuring a training run

The project configuration file can be found in conf/config.yaml.

Launching training

To launch a training session, simply run:

python scripts/train.py

Monitoring a training run

tensorboard --logdir experiments

Citing this repository

@software{Wilson_torchbox3d_2022,
  author = {Wilson, Benjamin and Pontes, Jhony},
  month = {4},
  title = {{torchbox3d}},
  url = {https://github.com/benjaminrwilson/torchbox3d},
  version = {0.0.1},
  year = {2022}
}

Acknowledgements

• https://github.com/mit-han-lab/torchsparse

@inproceedings{tang2020searching,
  title = {{Searching Efficient 3D Architectures with Sparse Point-Voxel Convolution}},
  author = {Tang, Haotian and Liu, Zhijian and Zhao, Shengyu and Lin, Yujun and Lin, Ji and Wang, Hanrui and Han, Song},
  booktitle = {European Conference on Computer Vision (ECCV)},
  year = {2020}
}