Research map / Point clouds / Point cloud classification
UAWTrack: Universal 3D Single Object Tracking in Adverse Weather
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes UAWTrack, a voxel-based, motion-centric 3D single object tracker with a mixture-of-experts weather-specific head, and builds KITTI-A and NuScenes-A adverse-weather benchmarks via physics-based simulation.
Problem
3D single object tracking in LiDAR point clouds degrades sharply under adverse weather (rain, snow, fog), which produces sparser and noisier point clouds, and no benchmarks existed for 3D SOT in such conditions.
Approach
Replace point-based Siamese features with voxel-based BEV features from two consecutive frames, model motion explicitly with multi-scale motion aggregation and motion-guided attention fusion, and route inputs to weather-specific experts via a top-K gating router trained on weather labels.
Key results
- On KITTI-A, UAWTrack reaches mean Success/Precision of 68.3/86.1 in clear weather and 60.3/74.9 in rain, outperforming MBPTrack (66.4/84.4 clear; 52.0/68.6 rain).
- On NuScenes-A, UAWTrack surpasses MBPTrack by 8.22% (Success) in clear weather and 3.03% in rain, per the paper's text.
- Ablation on KITTI-A shows the MFF module raises clear-weather mean Success from 64.3 to 71.8 for the car category.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add SP to fixed keypoints SF (ablation) | UAWTrack · KITTI-A Clear (Car) | Success: 70.2 -> 70.8 | Helped |
| Add SC to fixed keypoints SF (ablation) | UAWTrack · KITTI-A Clear (Car) | Success: 70.2 -> 71.1 | Helped |
| Add both SP and SC to SF (ablation) | UAWTrack · KITTI-A Clear (Car) | Success: 70.2 -> 71.8 | Helped |
| Add MFF (ablation) | UAWTrack · KITTI-A Clear (Car) | Success: 64.3 -> 71.8 | Helped |
| Use weather-specific experts instead of one shared expert (ablation) | UAWTrack · KITTI-A Clear (Car) | Success: 70.1 -> 71.8 | Helped |
| Use weather-specific experts instead of one shared expert (ablation) | UAWTrack · KITTI-A Rain (Car) | Success: 55.8 -> 58.3 | Helped |
| Use weather-specific experts instead of one shared expert (ablation) | UAWTrack · KITTI-A Snow (Car) | Success: 54.8 -> 57.2 | Helped |
| Use weather-specific experts instead of one shared expert (ablation) | UAWTrack · KITTI-A Fog (Car) | Success: 71.2 -> 72.8 | Helped |
| Use UAWTrack instead of MBPTrack | 3D SOT tracking · KITTI-A Clear | Success: 66.4 -> 68.3 | Helped |
| Use UAWTrack instead of MBPTrack | 3D SOT tracking · KITTI-A Rain | Success: 52.0 -> 60.3 | Helped |
| Use UAWTrack instead of MBPTrack | 3D SOT tracking · KITTI-A Snow | Success: 50.7 -> 57.0 | Helped |
| Use UAWTrack instead of MBPTrack | 3D SOT tracking · KITTI-A Fog | Success: 52.3 -> 57.7 | Helped |
Limitations
- Adverse-weather benchmarks are simulated rather than real recorded data, except that NuScenes-A is described as incorporating real-world conditions in the results section.
- Weather router relies on weather-type labels for training.
- Performance on pedestrian and van categories remains lower than on car in several weather conditions.
Experimental setup
Datasets: KITTI NuScenes KITTI-A NuScenes-A
Benchmarks: KITTI-A NuScenes-A
Metrics: Success Precision
Baselines: P2B M2Track GLT-T MBPTrack
Builds on: M2Track MBPTrack BAT GShard Sparsely-Gated MoE
Closest papers by meaning
- Modeling Continuous Motion for 3D Point Cloud Object Tracking (AAAI 2024 · similarity 0.88)
- An Empirical Study of the Generalization Ability of Lidar 3D Object Detectors to Unseen Domains (CVPR 2024 · similarity 0.87)
- M3SOT: Multi-Frame, Multi-Field, Multi-Space 3D Single Object Tracking (AAAI 2024 · similarity 0.87)
- Towards Robust 3D Object Detection with LiDAR and 4D Radar Fusion in Various Weather Conditions (CVPR 2024 · similarity 0.87)
- Sunshine to Rainstorm: Cross-Weather Knowledge Distillation for Robust 3D Object Detection (AAAI 2024 · similarity 0.86)
- L4DR: LiDAR-4DRadar Fusion for Weather-Robust 3D Object Detection (AAAI 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.