Research map / Detection and segmentation / Open-set and few-shot detection
OV3D-CG: Open-vocabulary 3D Instance Segmentation with Contextual Guidance
Mingquan Zhou, Chen He, Ruiping Wang, Xilin Chen
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 |
|---|---|---|---|
| represent instances with SAM mask rather than 2D patch crop (ablation) | OV3D-CG · ScanNet200 val | AP: 15.5 -> 22.7 | Improved |
| use bounding box context instead of patch (ablation) | OV3D-CG · ScanNet200 val | AP: 15.5 -> 20.3 | Improved |
| use SAM mask rather than landmarks (ablation) | OV3D-CG · ScanNet200 val | AP: 22.1 -> 22.7 | Improved |
| add chain-of-thought prompting to Gemini 1.5 Flash (ablation) | OV3D-CG · ScanNet200 val (SAI3D masks) | AP: 14.4 -> 17.7 | Improved |
| replace CLIP-style object crops with MLLM scene context (no CoT) (ablation) | OV3D-CG · ScanNet200 val (SAI3D masks) | APtail: 16.2 -> 9.8 | Hurt |
| use SigLIP2-base instead of CLIP (ablation) | OV3D-CG baseline · ScanNet200 val (SAI3D masks) | AP: 12.7 -> 11.8 | No clear effect |
| use Gemini 1.5 Flash instead of LLaVA-OneVision-7B (ablation) | OV3D-CG · ScanNet200 val (Mask3D masks) | AP: 15.7 -> 22.7 | Improved |
| use smaller Gemini 1.5 Flash-8B instead of Flash (ablation) | OV3D-CG · ScanNet200 val (Mask3D masks) | AP: 22.7 -> 19.7 | Hurt |
| combine 2D SAI3D-based and 3D Mask3D class-agnostic proposals | OV3D-CG · ScanNet200 | AP: 22.7 -> 25.4 | Improved |
| use MLLM contextual reasoning with oracle masks instead of CLIP (Open3DIS) | OV3D-CG · ScanNet200 (oracle masks) | AP: 30.9 -> 44.9 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- OpenMask3D — OpenMask3D: Open-Vocabulary 3D Instance Segmentation (NeurIPS 2023) · better on 2 of 2 shared results
- Open3DIS — Open3DIS: Open-Vocabulary 3D Instance Segmentation with 2D Mask Guidance (CVPR 2024) · better on 1 of 1 shared result
- SAI3D — SAI3D: Segment Any Instance in 3D Scenes (CVPR 2024) · 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
- Open-Vocabulary 3D Semantic Segmentation with Foundation Models (CVPR 2024 · similarity 0.93)
- Open3DIS: Open-Vocabulary 3D Instance Segmentation with 2D Mask Guidance (CVPR 2024 · similarity 0.92)
- Segment Any 3D Object with Language (ICLR 2025 · similarity 0.91)
- Mosaic3D: Foundation Dataset and Model for Open-Vocabulary 3D Segmentation (CVPR 2025 · similarity 0.89)
- COS3D: Collaborative Open-Vocabulary 3D Segmentation (NeurIPS 2025 · similarity 0.89)
- OpenLex3D: A Tiered Benchmark for Open-Vocabulary 3D Scene Representations (NeurIPS 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.
