[AAAI 2026] Redundant Queries in DETR-Based 3D Detection Methods: Unnecessary and Prunable

This is the implentation for our paper "Redundant Queries in DETR-Based 3D Detection Methods: Unnecessary and Prunable"

Introduction

We propose a simple yet effective gradual query pruning method that accelerates 3D object detection models.

News

• [2025/11/8]🔥Our paper is accepted by AAAI 2026 🎉🎉🎉.

Examples

We provide examples for PETR and StreamPETR.

Citation

@inproceedings{xu2025gpq,
      title={Redundant Queries in DETR-Based 3D Detection Methods: Unnecessary and Prunable}, 
      author={Lizhen Xu and Zehao Wu and Wenzhao Qiu and Shanmin Pang and Xiuxiu Bai and Kuizhi Mei and Jianru Xue},
      booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
      year={2025},
}