Research map / Detection and segmentation / 3D detection and driving
STUR3D: Spatio-Temporal Unified Representation Learning for 3D Object Detection
Huijie Fan, Pengrui Huang, Qiang Wang, Baojie Fan, Jiahua Dong, Liangqiong Qu
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 STOPP temporal prior propagator (ablation) | STUR3D baseline · nuScenes val ablation (V2-99 320x800) | mAP (%): 48.2 -> 48.9 | Improved |
| add STGE geometry encoder (ablation) | STUR3D baseline · nuScenes val ablation (V2-99 320x800) | NDS (%): 57.1 -> 58.0 | Improved |
| add OQG object-center query generator (ablation) | STUR3D baseline · nuScenes val ablation (V2-99 320x800) | mAP (%): 48.2 -> 50.2 | Improved |
| combine STOPP and OQG (ablation) | STUR3D baseline · nuScenes val ablation (V2-99 320x800) | mAP (%): 48.2 -> 52.1 | Improved |
| add all three modules (ablation) | STUR3D baseline · nuScenes val ablation (V2-99 320x800) | mAP (%): 48.2 -> 53.0 | Improved |
| remove 2D detections from STOPP priors (ablation) | STUR3D · nuScenes val | mAP (%): 53.0 -> 52.1 | Hurt |
| remove 3D detections from STOPP priors (ablation) | STUR3D · nuScenes val | mAP (%): 53.0 -> 52.4 | Hurt |
| increase cached frames from one to two (ablation) | STOPP · nuScenes val | mAP (%): 53.0 -> 52.9 | No clear effect |
| increase cached frames from one to four (ablation) | STOPP · nuScenes val | mAP (%): 53.0 -> 52.3 | Hurt |
| increase 2D detections to 256 or reduce to 64 (ablation) | STOPP · nuScenes val | mAP (%): 53.0 -> 52.6 (256); 52.7 (64 with 128 3D) | Hurt |
| replace STGE depth encoding with Linear or MLP (ablation) | STUR3D · nuScenes val | mAP (%): 53.0 -> 52.7 (Linear); 52.0 (MLP) | Hurt |
| add LiDAR depth supervision during training (ablation) | STGE · nuScenes val | NDS (%): 60.2 -> 61.2 | Improved |
Closest papers by meaning
- LocateAnything3D: Vision-Language 3D Detection with Chain-of-Sight (CVPR 2026 · similarity 0.88)
- Track3R: Joint Point Map and Trajectory Prior for Spatiotemporal 3D Understanding (NeurIPS 2025 · similarity 0.88)
- OV-Uni3DETR: Towards Unified Open-Vocabulary 3D Object Detection via Cycle-Modality Propagation (ECCV 2024 · similarity 0.88)
- ImageNet3D: Towards General-Purpose Object-Level 3D Understanding (NeurIPS 2024 · similarity 0.87)
- Improving Distant 3D Object Detection Using 2D Box Supervision (CVPR 2024 · similarity 0.86)
- VLM-3R: Vision-Language Models Augmented with Instruction-Aligned 3D Reconstruction (CVPR 2026 · 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.
