Home

Awesome

<div align="center"> <h1>[CVPR 2024] Enhancing 3D Object Detection with 2D Detection-Guided Query Anchors</h1> </div> <div align="center"> <img src="figs/framework.png" width="800"/> </div><br/>

Introduction

This repository is the official implementation of our paper "Enhancing 3D Object Detection with 2D Detection-Guided Query Anchors, CVPR 2024 arXiv". Our code is based on StreamPETR.

Getting Started

Please follow the docs below.

  1. Environment Setup.
  2. Data Preparation.
  3. Inference.

Results on NuScenes Val Set.

MethodsBackboneImage SizeNDSmAPconfigmodel
StreamPETRV2-99320×80057.148.2--
StreamPETR-QAF2D (Ours)V2-99320×80058.849.8configmodel

Comparison of the base detectors and their QAF2D enhanced version on the nuScenes validation split.

Note

Due to some internal policies, we do not release the full codebase, and the current 2D detection results are read from a saved file.

Citation

If you find QAF2D useful in your research or applications, please consider citing it. Thank you.

@inproceedings{ji2024enhancing,
  title={Enhancing 3D Object Detection with 2D Detection-Guided Query Anchors},
  author={Ji, Haoxuanye and Liang, Pengpeng and Cheng, Erkang},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and 
  Pattern Recognition},
  pages={21178--21187},
  year={2024}
}