• Label-Guided Auxiliary Training Improves 3D Object Detector
• • Introduction
  • Citation
  • Installation
  • Datasets
  • • ScanNet
    • SUN RGB-D
  • Training and testing
  • • Training
    • Testing mAP
    • visualization
  • Acknowledgement
  • Code Modifications
  • License Compliance

Label-Guided Auxiliary Training Improves 3D Object Detector

This is a release of the code of our paper Label-Guided Auxiliary Training Improves 3D Object Detector, ECCV 2022.

Authors: Yaomin Huang, Xinmei Liu, Yichen Zhu, Zhiyuan Xu, Chaomin Shen *, Zhengping Che, Guixu Zhang,
Yaxin Peng, Feifei Feng, and Jian Tang * (*corresponding author)

[arxiv]

In this repository, we reimplement LG3D based on mmdetection3d for easier usage.

Introduction

In this
paper, we propose a Label-Guided auxiliary training method for 3D object detection (LG3D in short), which serves as an auxiliary network to enhance the feature learning of existing 3D object detectors.

Citation

If you find our work useful in your research, please consider citing:

@inproceedings{huang2022label,
  title={Label-guided auxiliary training improves 3d object detector},
  author={Huang, Yaomin and Liu, Xinmei and Zhu, Yichen and Xu, Zhiyuan and Shen, Chaomin and Che, Zhengping and Zhang, Guixu and Peng, Yaxin and Feng, Feifei and Tang, Jian},
  booktitle={Computer Vision--ECCV 2022: 17th European Conference, Tel Aviv, Israel, October 23--27, 2022, Proceedings, Part IX},
  pages={684--700},
  year={2022},
  organization={Springer}
}

Installation

This repo is built based on mmdetection3d (V1.0.0), please follow the getting_started.md for installation.

The code is tested under the following environment:

• Ubuntu 18.04 LTS
• Python 3.7.16
• Pytorch 1.6.0
• CUDA 10.2
• GCC 7.5.0
• mmcv-full 1.5.2
• mmdet 2.24.0
• mmsegmentation 0.29.0

Datasets

ScanNet

Please follow the instruction here to prepare ScanNet Data.

SUN RGB-D

Please follow the instruction here to prepare SUN RGB-D Data.

Training and testing

Training

For ScanNet V2, please run to training LG3D:

./tools/dist_train.sh configs/votenet/votenet_lg3d_scannet.py 4 --work-dir ./log/scannet/lg3d/

After training, please run :

python .tools/convert_fully2single.py

convert fully LG3D model to VoteNet.

Testing mAP

Test VoteNet on ScanNet and evaluate the mAP.

python tools/test.py configs/votenet/votenet_8x8_scannet-3d-18class.py log/pre_trained/votenet_scannet_final.pth --eval mAP --eval-options 'out_dir=./log/scannet/show_results'

visualization

Test VoteNet on ScanNet and save the points and prediction visualization results.

python tools/test.py configs/votenet/votenet_8x8_scannet-3d-18class.py log/pre_trained/votenet_scannet_final.pth --show --show-dir ./log/scannet/show_results

Acknowledgement

Our code is heavily based on MMDetection3D. Thanks mmdetection3d Development Team for their awesome codebase.

Code Modifications

The following modifications have been made to the original code:

• Added LG3D configuration file: configs/votenet/votenet_lg3d_scannet.py
• Added LG3D model file: mmdet3d/model/detector/lg3d_votenet.py
• Added LG3D required corresponding module code: mmdet3d/models/backbone/attention_utils.py

License Compliance

Please ensure compliance with the licensing terms of the original project, i.e., the Apache License.