Research map / Detection and segmentation / 3D detection and driving
HyperDet3D: Learning a Scene-Conditioned 3D Object Detector
Yu Zheng, Yueqi Duan, Jiwen Lu, Jie Zhou, Qi Tian
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 scene-conditioned hypernetworks with MSA and DDH | GF3D-level 3D detection on ScanNet V2 · ScanNet V2 validation | mAP@0.5: 52.8 -> 57.2 | Improved |
| add scene-conditioned hypernetworks with MSA and DDH | GF3D-level 3D detection on SUN RGB-D V1 · SUN RGB-D V1 validation | mAP@0.5: 45.2 -> 47.3 | Improved |
| add scene-conditioned knowledge using SSA (ablation) | DDH-only baseline · ScanNet V2 validation | mAP@0.5: 53.5 -> 54.7 | Improved |
| replace SSA with multi-head MSA (ablation) | HyperDet3D with scene-agnostic and scene-specific knowledge · ScanNet V2 validation | mAP@0.5: +2.5% | Improved |
| use scene-specific knowledge without scene-agnostic knowledge (ablation) | HyperDet3D · ScanNet V2 validation | mAP@0.5: 57.2 -> 55.4 | Hurt |
| use scene-agnostic knowledge without scene-specific knowledge (ablation) | HyperDet3D · ScanNet V2 validation | mAP@0.5: 57.2 -> 53.8 | Hurt |
| add the disentangled detection head (ablation) | MSA with scene-agnostic and scene-specific knowledge · ScanNet V2 validation | mAP@0.5: 56.8 -> 57.2 | Improved |
| add a scene-type classification branch | GF3D · ScanNet V2 validation | mAP@0.5: 52.8 -> 54.3 | Improved |
| use HyperDet3D rather than GF3D | cross-dataset detector pretrained on SUN RGB-D V1 · Cross-dataset ScanNet V2 validation | mAP8: 35.2 -> 43.6 | Improved |
| remove scene-specific knowledge (ablation) | HyperDet3D in cross-dataset evaluation · Cross-dataset ScanNet V2 validation | mAP8: -3.0% | Hurt |
| replace MSA with SSA (ablation) | HyperDet3D in cross-dataset evaluation · Cross-dataset ScanNet V2 validation | mAP8: -3.4% | Hurt |
| add scene-conditioned detection | GF3D on SUN RGB-D V1 bathtub detections · SUN RGB-D V1 validation, bathtub | AP: 64.0 -> 59.1 | Hurt |
Closest papers by meaning
- Learning Dynamic Scene-Conditioned 3D Object Detectors (TPAMI 2023 · similarity 0.96)
- Disentangling 3D Prototypical Networks for Few-Shot Concept Learning (ICLR 2021 · similarity 0.87)
- Learning 3D Scene Priors With 2D Supervision (CVPR 2023 · similarity 0.87)
- SemLT3D: Semantic-Guided Expert Distillation for Camera-only Long-Tailed 3D Object Detection (CVPR 2026 · similarity 0.87)
- Towards 3D Objectness Learning in an Open World (NeurIPS 2025 · similarity 0.86)
- Uncertainty Meets Diversity: A Comprehensive Active Learning Framework for Indoor 3D Object Detection (CVPR 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.
