Research map / Point clouds / Point cloud classification
RegTrack: Simplicity Beneath Complexity in Robust Multi-Modal 3D Multi-Object Tracking
Lipeng Gu, Xuefeng Yan, Song Wang, Mingqiang Wei
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add image cue during training (ablation) | RegTrack point-cloud encoder · nuScenes val | AMOTA: 70.0 -> 72.0 | Improved |
| add MoE-GEnc geometry cue (ablation) | RegTrack UTEnc · nuScenes val | AMOTA: 72.0 -> 73.8 | Improved |
| add image cue at inference (ablation) | RegTrack · nuScenes val | AMOTA: 73.8 -> 73.5 | Hurt |
| use sinusoidal instead of MLP positional encoding (ablation) | LG-PEnc · nuScenes val | AMOTA: 71.1 -> 73.6 | Improved |
| drop z channel, keep only x and y (ablation) | LG-PEnc · nuScenes val | AMOTA: 73.6 -> 73.8 | Improved |
| use two experts instead of one (ablation) | MoE-GEnc · nuScenes val | AMOTA: 73.3 -> 73.8 | Improved |
| use three experts instead of two (ablation) | MoE-GEnc · nuScenes val | AMOTA: 73.8 -> 73.6 | No clear effect |
| add global features to local features (ablation) | LG-PEnc · nuScenes val | AMOTA: 73.1 -> 73.6 | Improved |
| add hybrid attention (ablation) | LG-PEnc · nuScenes val | AMOTA: 73.6 -> 73.8 | Improved |
| use channel-wise instead of point-wise attention normalization (ablation) | LG-PEnc · nuScenes val | Recall: +0.9% Recall, +210 TP | Improved |
| use adaptive instead of fixed local-global aggregation (ablation) | LG-PEnc · nuScenes val | AMOTA: 73.0 -> 73.8 | Improved |
| raise fixed association threshold from 0.5 to 0.9 (ablation) | RegTrack · nuScenes val | AMOTA: 73.8 -> 64.5 | Hurt |
Closest papers by meaning
- M3SOT: Multi-Frame, Multi-Field, Multi-Space 3D Single Object Tracking (AAAI 2024 · similarity 0.89)
- SpatialTrackerV2: Advancing 3D Point Tracking with Explicit Camera Motion (ICCV 2025 · similarity 0.89)
- TrackCraft3R: Repurposing Video Diffusion Transformers for Dense 3D Tracking (NeurIPS 2026 · similarity 0.89)
- Generalizable Structure-Aware Keypoint Correspondence for Category-Unified 3D Single Object Tracking (CVPR 2026 · similarity 0.89)
- 3DMOTFormer: Graph Transformer for Online 3D Multi-Object Tracking (ICCV 2023 · similarity 0.89)
- DiffusionTrack: Diffusion Model for Multi-Object Tracking (AAAI 2024 · similarity 0.88)
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.
