Research map / Detection and segmentation / 3D detection and driving
OpenM3D: Open Vocabulary Multi-view Indoor 3D Object Detection without Human Annotations
Peng-Hao Hsu, Ke Zhang, Fu-En Wang, Tao Tu, Ming-Feng Li, Yu-Lun Liu, Albert Y. C. Chen, Min Sun, Cheng-Hao Kuo
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 |
|---|---|---|---|
| use graph-embedding pseudo boxes instead of OV-3DET boxes for training (ablation) | OpenM3D · ScanNet200 class-agnostic | AP@25 (%): 19.53 -> 26.92 | Improved |
| use graph-embedding pseudo boxes instead of SAM3D boxes for training (ablation) | OpenM3D · ScanNet200 OV detection | mAP@25 (%): 3.92 -> 4.23 | Improved |
| add mesh segmentation refinement to pseudo boxes (ablation) | OpenM3D · ScanNet200 class-agnostic | AP@25 (%): 25.95 -> 26.92 | Improved |
| use CropFormer instead of SAM for 2D segments (ablation) | OpenM3D · ScanNet200 OV detection | mAP@25 (%): 4.23 -> 4.76 mAP; 15.12 -> 14.62 mAR | Mixed |
| use single-stage voxel-semantic alignment instead of two-stage CLIP classification (S2D) (ablation) | OpenM3D · ScanNet200 OV detection | mAR@25 (%): 10.05 -> 15.12 | Improved |
| use single-stage OpenM3D instead of S2D (ablation) | OpenM3D · ScanNet200 OV detection | mAP@25 (%): 4.17 -> 4.23 | No clear effect |
| use single-stage OpenM3D instead of S2D (ablation) | OpenM3D · ARKitScenes | mAR@25: 19.58 -> 42.77 | Improved |
| swap CLIP ViT-L/14 for ViT-B/16 or ViT-B/32 in alignment (ablation) | OpenM3D · ScanNet200 OV detection | mAP@25 (%): 4.23 -> 4.16 (B/16), 4.02 (B/32) | No clear effect |
| use CIFAR100 prompts instead of ImageNet prompts (ablation) | OpenM3D · ScanNet200 OV detection | mAP@25 (%): 4.23 -> 4.14 | No clear effect |
| use graph embedding clustering of 2D segments instead of SAM3D sequential merging (ablation) | 3D pseudo box generation · ScanNet200 pseudo boxes | precision@25 (%): 14.48 -> 32.07 | Improved |
| pseudo boxes vs SAM3D on recall@50 (ablation) | 3D pseudo box generation · ScanNet200 pseudo boxes | recall@50 (%): 36.07 -> 32.99 | Hurt |
| use estimated depth with multi-view depth estimator instead of OpenM3D | S2D baseline · ScanNet200 inference time | seconds per scene: 0.3 -> 81 | Hurt |
Later papers that compared with this method
2 later accepted papers report OpenM3D as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Zoo3D: Zero-Shot 3D Object Detection at Scene Level (CVPR 2026 · better on 3 of 3 shared results)
- Group3D: MLLM-Guided Semantic Grouping for Open-Vocabulary 3D Object Detection (ECCV 2026 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- ImOV3D: Learning Open Vocabulary Point Clouds 3D Object Detection from Only 2D Images (NeurIPS 2024 · similarity 0.94)
- Training an Open-Vocabulary Monocular 3D Detection Model without 3D Data (NeurIPS 2024 · similarity 0.90)
- Open3DIS: Open-Vocabulary 3D Instance Segmentation with 2D Mask Guidance (CVPR 2024 · similarity 0.89)
- Rooms from Motion: Un-posed Indoor 3D Object Detection as Localization and Mapping (NeurIPS 2025 · similarity 0.89)
- Hierarchical Cross-Modal Alignment for Open-Vocabulary 3D Object Detection (AAAI 2025 · similarity 0.89)
- Cubify Anything: Scaling Indoor 3D Object Detection (CVPR 2025 · similarity 0.89)
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.
