Monocular Quasi-Dense 3D Object Tracking (QD-3DT) is an online framework detects and tracks objects in 3D using quasi-dense object proposals from 2D images.
Monocular Quasi-Dense 3D Object Tracking,
Hou-Ning Hu, Yung-Hsu Yang, Tobias Fischer, Trevor Darrell, Fisher Yu, Min Sun,
Paper (arXiv 2103.07351) Project Website (QD-3DT)
@article{hu2022monocular,
title={Monocular quasi-dense 3d object tracking},
author={Hu, Hou-Ning and Yang, Yung-Hsu and Fischer, Tobias and Darrell, Trevor and Yu, Fisher and Sun, Min},
journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
year={2022},
publisher={IEEE}
}
[2024/05/16]: We released the code for our CC-3DT paper. Please check the here for more details and enjoy new tracking performance!
A reliable and accurate 3D tracking framework is essential for predicting future locations of surrounding objects and planning the observer’s actions in numerous applications such as autonomous driving. We propose a framework that can effectively associate moving objects over time and estimate their full 3D bounding box information from a sequence of 2D images captured on a moving platform. The object association leverages quasi-dense similarity learning to identify objects in various poses and viewpoints with appearance cues only. After initial 2D association, we further utilize 3D bounding boxes depth-ordering heuristics for robust instance association and motion-based 3D trajectory prediction for re-identification of occluded vehicles. In the end, an LSTM-based object velocity learning module aggregates the long-term trajectory information for more accurate motion extrapolation. Experiments on our proposed simulation data and real-world benchmarks, including KITTI, nuScenes, and Waymo datasets, show that our tracking framework offers robust object association and tracking on urban-driving scenarios. On the Waymo Open benchmark, we establish the first camera-only baseline in the 3D tracking and 3D detection challenges. Our quasi-dense 3D tracking pipeline achieves impressive improvements on the nuScenes 3D tracking benchmark with near five times tracking accuracy of the best vision-only submission among all published methods.
We achieved the best vision-only submission
| AMOTA | AMOTP |
|---|---|
| 21.7 | 1.55 |
We established the first camera-only baseline on Waymo Open
| MOTA/L2 | MOTP/L2 |
|---|---|
| 0.0001 | 0.0658 |
| MOTA | MOTP |
|---|---|
| 86.44 | 85.82 |
Please refer to INSTALL.md for installation and to DATA.md dataset preparation.
Please see GETTING_STARTED.md for the basic usage of QD-3DT.
Please refer to MODEL_ZOO.md for reproducing the results on varients of benchmarks
This repo is currently maintained by Hou-Ning Hu (@eborboihuc), Yung-Hsu Yang (@RoyYang0714), and Tobias Fischer (@tobiasfshr).
This work is licensed under BSD 3-Clause License. See LICENSE for details. Third-party datasets and tools are subject to their respective licenses.
We thank Jiangmiao Pang for his help in providing the qdtrack codebase in mmdetection. This repo uses py-motmetrics for MOT evaluation, waymo-open-dataset for Waymo Open 3D detection and 3D tracking task, and nuscenes-devkit for nuScenes evaluation and preprocessing.