MonoATT: Online Monocular 3D Object Detection with Adaptive Token Transformer

被引:5
|
作者
Zhou, Yunsong [1 ]
Zhu, Hongzi [1 ]
Liu, Quan [1 ]
Chang, Shan [2 ]
Guo, Minyi [1 ]
机构
[1] Shanghai Jiao Tong Univ, Shanghai, Peoples R China
[2] Donghua Univ, Shanghai, Peoples R China
基金
中国国家自然科学基金; 上海市自然科学基金;
关键词
D O I
10.1109/CVPR52729.2023.01678
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Mobile monocular 3D object detection (Mono3D) (e.g., on a vehicle, a drone, or a robot) is an important yet challenging task. Existing transformer-based offline Mono3D models adopt grid-based vision tokens, which is suboptimal when using coarse tokens due to the limited available computational power. In this paper, we propose an online Mono3D framework, called MonoATT, which leverages a novel vision transformer with heterogeneous tokens of varying shapes and sizes to facilitate mobile Mono3D. The core idea of MonoATT is to adaptively assign finer tokens to areas of more significance before utilizing a transformer to enhance Mono3D. To this end, we first use prior knowledge to design a scoring network for selecting the most important areas of the image, and then propose a token clustering and merging network with an attention mechanism to gradually merge tokens around the selected areas in multiple stages. Finally, a pixel-level feature map is reconstructed from heterogeneous tokens before employing a SOTA Mono3D detector as the underlying detection core. Experiment results on the real-world KITTI dataset demonstrate that MonoATT can effectively improve the Mono3D accuracy for both near and far objects and guarantee low latency. MonoATT yields the best performance compared with the state-of-the-art methods by a large margin and is ranked number one on the KITTI 3D benchmark.
引用
收藏
页码:17493 / 17503
页数:11
相关论文
共 50 条
  • [41] MonoSample: Synthetic 3D Data Augmentation Method in Monocular 3D Object Detection
    Qiao, Junchao
    Liu, Biao
    Yang, Jiaqi
    Wang, Baohua
    Xiu, Sanmu
    Du, Xin
    Nie, Xiaobo
    [J]. IEEE ROBOTICS AND AUTOMATION LETTERS, 2024, 9 (08): : 7326 - 7332
  • [42] Understanding Depth Map Progressively: Adaptive Distance Interval Separation for Monocular 3d Object Detection
    Cheng, Xianhui
    Qiu, Shoumeng
    Zou, Zhikang
    Pu, Jian
    Xue, Xiangyang
    [J]. 2023 INTERNATIONAL JOINT CONFERENCE ON NEURAL NETWORKS, IJCNN, 2023,
  • [43] Long-Short Range Adaptive Transformer With Dynamic Sampling for 3D Object Detection
    Wang, Chuxin
    Deng, Jiacheng
    He, Jianfeng
    Zhang, Tianzhu
    Zhang, Zhe
    Zhang, Yongdong
    [J]. IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, 2023, 33 (12) : 7616 - 7629
  • [44] GAC3D: improving monocular 3D object detection with ground-guide model and adaptive convolution
    Bui, Minh-Quan Viet
    Ngo, Duc Tuan
    Pham, Hoang-Anh
    Nguyen, Duc Dung
    [J]. PEERJ COMPUTER SCIENCE, 2021, 7
  • [45] eGAC3D: enhancing depth adaptive convolution and depth estimation for monocular 3D object pose detection
    Ngo, Duc Tuan
    Bui, Minh-Quan Viet
    Nguyen, Duc Dung
    Pham, Hoang-Anh
    [J]. PEERJ COMPUTER SCIENCE, 2022, 8
  • [46] MonoRUn: Monocular 3D Object Detection by Reconstruction and Uncertainty Propagation
    Chen, Hansheng
    Huang, Yuyao
    Tian, Wei
    Gao, Zhong
    Xiong, Lu
    [J]. 2021 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, CVPR 2021, 2021, : 10374 - 10383
  • [47] Monocular 3D Object Detection: An Extrinsic Parameter Free Approach
    Zhou, Yunsong
    He, Yuan
    Zhu, Hongzi
    Wang, Cheng
    Li, Hongyang
    Jiang, Qinhong
    [J]. 2021 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, CVPR 2021, 2021, : 7552 - 7562
  • [48] Monocular 3D Object Detection Based on Uncertainty Prediction of Keypoints
    Chen, Mu
    Zhao, Huaici
    Liu, Pengfei
    [J]. MACHINES, 2022, 10 (01)
  • [49] Efficient Active Learning Strategies for Monocular 3D Object Detection
    Hekimoglu, Aral
    Schmidt, Michael
    Marcos-Ramiro, Alvaro
    Rigoll, Gerhard
    [J]. 2022 IEEE INTELLIGENT VEHICLES SYMPOSIUM (IV), 2022, : 295 - 302
  • [50] MonoDCN: Monocular 3D object detection based on dynamic convolution
    Qu, Shenming
    Yang, Xinyu
    Gao, Yiming
    Liang, Shengbin
    [J]. PLOS ONE, 2022, 17 (10):