Research map / Detection and segmentation / 3D detection and driving
Time3D: End-to-End Joint Monocular 3D Object Detection and Tracking for Autonomous Driving
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| jointly train detection and tracking end-to-end rather than train them separately | Time3D on nuScenes test · nuScenes test | AMOTA: 15.6 -> 21.4; improves 37% | Helped |
| add spatial-temporal information flow instead of MLP layers and cosine similarity (ablation) | Time3D · nuScenes val | AMOTA: 17.9 -> 26.0; improves by 45% | Helped |
| add spatial information flow to the detector/tracker baseline (ablation) | Time3D · nuScenes val | AMOTA: 17.9 -> 23.9 | Helped |
| add temporal information flow after spatial flow (ablation) | Time3D · nuScenes val | AMOTA: 23.9 -> 26.0 | Helped |
| add 3D box geometry to Re-ID cues (ablation) | Time3D heterogeneous cue embedding · nuScenes val | AMOTA: 16.5 -> 24.7 | Helped |
| add 2D box geometry to 3D box and Re-ID cues (ablation) | Time3D heterogeneous cue embedding · nuScenes val | AMOTA: 24.7 -> 25.8 | Helped |
| add category cue to 2D box, 3D box, and Re-ID cues (ablation) | Time3D heterogeneous cue embedding · nuScenes val | AMOTA: 25.8 -> 26.0 | Helped |
| encode box geometry through corners rather than directly encode box parameters (ablation) | Time3D geometric embedding · nuScenes val | AMOTA: 19.0 -> 26.0 | Helped |
| add Re-ID features to the detection model (ablation) | Time3D detector · nuScenes val | mAP: 32.2 -> 29.1 | Hurt |
| add Re-ID features to the detection model (ablation) | Time3D detector · nuScenes val | NDS: 39.5 -> 39.0 | Hurt |
| add temporal-consistency loss to constrain relative 3D geometry across frames (ablation) | Time3D · nuScenes | trajectory smoothness | Helped |
| use end-to-end joint detection and tracking rather than separate stages | Time3D on nuScenes test · nuScenes test | MOTA: 14.1 -> 17.3; improves 22% | Helped |
Closest papers by meaning
- M3SOT: Multi-Frame, Multi-Field, Multi-Space 3D Single Object Tracking (AAAI 2024 · similarity 0.89)
- TrackCraft3R: Repurposing Video Diffusion Transformers for Dense 3D Tracking (NeurIPS 2026 · similarity 0.89)
- ADA-Track: End-to-End Multi-Camera 3D Multi-Object Tracking with Alternating Detection and Association (CVPR 2024 · similarity 0.89)
- Leveraging Temporal Cues for Semi-Supervised Multi-View 3D Object Detection (CVPR 2025 · similarity 0.88)
- S2-Track: A Simple yet Strong Approach for End-to-End 3D Multi-Object Tracking (ICML 2025 · similarity 0.88)
- TrackingWorld: World-centric Monocular 3D Tracking of Almost All Pixels (NeurIPS 2025 · 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.