Research map / Detection and segmentation / 3D detection and driving
Group3D: MLLM-Guided Semantic Grouping for Open-Vocabulary 3D Object Detection
Youbin Kim, Jinho Park, Hogun Park, Eunbyung Park
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 semantic compatibility grouping instead of geometry-only merging without category (ablation) | Group3D fragment merging · ScanNet20 (pose-free) | mAP25: 28.2 -> 41.2 | Improved |
| use semantic compatibility groups instead of strict same-category constraint (ablation) | Group3D fragment merging · ScanNet20 (pose-free) | mAP25: 35.9 -> 41.2 | Improved |
| add same-category gating over geometry-only merging (ablation) | Group3D fragment merging · ScanNet20 (pose-free) | mAP25: 28.2 -> 35.9 | Improved |
| replace GPT-5.1 with smaller Qwen3-VL-8B as MLLM (ablation) | Group3D · ScanNet20 (pose-free) | mAP25: 41.2 -> 38.5 | Hurt |
| replace SAM 3 with Grounded SAM 2 (ablation) | Group3D · ScanNet20 (pose-free) | mAP25: 41.2 -> 39.7 | Hurt |
| replace Depth Anything 3 with VGGT for reconstruction (ablation) | Group3D · ScanNet20 (pose-free) | mAP25: 41.2 -> 40.0 | Mixed |
| increase category hypotheses per view from K=5 to K=10 (ablation) | Group3D · ScanNet20 (pose-free) | mAP50: 18.5 -> 18.8 | No clear effect |
| use 10 cm voxels instead of 5 cm (ablation) | Group3D geometric overlap · ScanNet20 (pose-free) | mAP25: 41.2 -> 37.9 | Hurt |
| use 1 cm voxels instead of 5 cm (ablation) | Group3D geometric overlap · ScanNet20 (pose-free) | mAP25: 41.2 -> 41.3 | No clear effect |
| increase number of input frames from 32 to 128 (ablation) | Group3D · ScanNet20 (pose-free) | mAP25: 36.2 -> 41.2 | Improved |
| use known camera poses instead of pose-free reconstruction | Group3D · ScanNet20 | mAP25: 41.2 -> 51.1 | Improved |
| Group3D versus Zoo3D_1 in pose-free zero-shot setting | multi-view open-vocabulary detection · ScanNet20 (pose-free) | mAP25: 27.9 -> 41.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Zoo3D_1 — Zoo3D: Zero-Shot 3D Object Detection at Scene Level (CVPR 2026) · better on 12 of 12 shared results
- OpenM3D — OpenM3D: Open Vocabulary Multi-view Indoor 3D Object Detection without Human Annotations (ICCV 2025) · 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
- Open3DIS: Open-Vocabulary 3D Instance Segmentation with 2D Mask Guidance (CVPR 2024 · similarity 0.90)
- Segment Any 3D Object with Language (ICLR 2025 · similarity 0.90)
- Towards 3D Objectness Learning in an Open World (NeurIPS 2025 · similarity 0.90)
- ImOV3D: Learning Open Vocabulary Point Clouds 3D Object Detection from Only 2D Images (NeurIPS 2024 · similarity 0.90)
- GOV-NeSF: Generalizable Open-Vocabulary Neural Semantic Fields (CVPR 2024 · similarity 0.89)
- ImageNet3D: Towards General-Purpose Object-Level 3D Understanding (NeurIPS 2024 · 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.
