Research map / Detection and segmentation / Open-set and few-shot detection
Find Any Part in 3D
Ziqi Ma, Yisong Yue, Georgia Gkioxari
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 |
|---|---|---|---|
| evaluate zero-shot Find3D versus best baseline PointCLIPV2 on unseen categories | PointCLIPV2 · Objaverse-General Unseen | mIoU: 10.27 -> 26.21 (260% improvement in mIoU) | Improved |
| change prompt from top-k to {part} of a {object} for PointCLIPV2 | PointCLIPV2 · ShapeNet-Part | mIoU: 48.47 -> 17.42 (64% decrease) | Hurt |
| random rotation of test objects for PointCLIPV2 | PointCLIPV2 · ShapeNet-Part | mIoU: 46% drop | Hurt |
| random rotation of test objects for Find3D | Find3D · ShapeNet-Part | mIoU: 28.39 -> 29.64 | Improved |
| domain shift to ShapeNetPart-V2 for PointCLIPV2 | PointCLIPV2 · ShapeNetPart-V2 | mIoU: 48.47 to 21.18 (56% drop) | Hurt |
| domain shift to ShapeNetPart-V2 for Find3D | Find3D · ShapeNetPart-V2 | mIoU: 20% increase | Improved |
| domain shift to ShapeNetPart-V2 for closed-world PointNeXt | PointNeXt · ShapeNetPart-V2 | mIoU: 80.44 -> 28.70 (64% drop) | Hurt |
| vary query prompt for PartSLIP++ on PartNet-E | PartSLIP++ · PartNet-E | mIoU: 32.71 -> 5.12 (84% drop) | Hurt |
| scale training object categories from 16 to 761 with the data engine (ablation) | Find3D · Objaverse-General Unseen | mIoU | Improved |
| use top-k prompt search instead of direct prompts | PointCLIPV2 vs Find3D · ShapeNet-Part canonical | mIoU: PointCLIPV2 48.666 vs Find3D 43.613 | Mixed |
| Find3D zero-shot vs PartSLIP++ trained on PartNet-E with part prompt | PartSLIP++ · PartNet-E | mIoU: 32.71 vs 16.38 | Hurt |
| Find3D feed-forward vs PartSLIP++ 2D-3D aggregation | PartSLIP++ · PartNet-E subset | inference time: 174.3s -> 0.9s | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- OpenMask3D — OpenMask3D: Open-Vocabulary 3D Instance Segmentation (NeurIPS 2023) · better on 5 of 5 shared results
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 0.86)
- PartNeXt: A Next-Generation Dataset for Fine-Grained and Hierarchical 3D Part Understanding (NeurIPS 2025 · similarity 0.86)
- 3DCoMPaT200: Language Grounded Large-Scale 3D Vision Dataset for Compositional Recognition (NeurIPS 2024 · similarity 0.86)
- PartDistill: 3D Shape Part Segmentation by Vision-Language Model Distillation (CVPR 2024 · similarity 0.85)
- Mosaic3D: Foundation Dataset and Model for Open-Vocabulary 3D Segmentation (CVPR 2025 · similarity 0.85)
- Segment Any 3D Object with Language (ICLR 2025 · similarity 0.85)
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.
