Skip to content

hongxiaoy/OV-3DET

 
 

Repository files navigation

OV-3DET: Open-Vocabulary Point-Cloud Object Detection without 3D Annotation

OV-3DET: An Open Vocabulary 3D DETector.

Paper | BibTeX

OV-3DET: Open-Vocabulary Point-Cloud Object Detection without 3D Annotation,
Yuheng Lu, Chenfeng Xu, Xiaobao Wei, Xiaodong Xie, Masayoshi Tomizuka, Kurt Keutzer and Shanghang Zhang,
Accepted to CVPR2023

Features

  • Detects 3D objects according to text prompting.

  • The training of OV-3DET does not require 3D annotation.

Installation

See installation instructions.

Dataset preparation

See dataset instructions, or directly download the processed dataset.

Training OV-3DET

Phase 1

Learn to Localize 3D Objects from 2D Pretrained Detector:

bash scripts/scannet_train_loc.sh

Phase 2

Learn to Classify 3D Objects from 2D Pretrained vision-language Model:

bash scripts/scannet_train_dtcc.sh

Evaluate OV-3DET

To evaluate OV-3DET, simply by running:

bash scripts/evaluate.sh

Pretrained Models

We provide the pretrained model weights for both "Phase 1" and "Phase 2".

Dataset Phase Epochs Model weights
ScanNet 1 400 weights
ScanNet 2 50 weights
SUN RGB-D 1 400 weights
SUN RGB-D 2 50 weights

Acknowledgement

This codebase is modified base on 3DETR [1], CLIP [2] and Detic [3], we sincerely appreciate their contributions!

[1] An end-to-end transformer model for 3d object detection. ICCV. 2021.
[2] Learning transferable visual models from natural language supervision. ICML. 2021.
[3] Detecting twenty-thousand classes using image-level supervision. ECCV. 2022.

Citation

If you find this repository helpful, please consider citing our work:

@article{lu2023open,
  title={Open-Vocabulary Point-Cloud Object Detection without 3D Annotation},
  author={Lu, Yuheng and Xu, Chenfeng and Wei, Xiaobao and Xie, Xiaodong and Tomizuka, Masayoshi and Keutzer, Kurt and Zhang, Shanghang},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year={2023}
}

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 96.2%
  • Cython 2.6%
  • Shell 1.2%