Research map / Point clouds / Segmentation and detection
GeoPurify: A Data-Efficient Geometric Distillation Framework for Open-Vocabulary 3D Segmentation
Weijia Dou, Xu Zhang, Yi Bin, Jian Liu, Bo Peng, Guoqing Wang, Yang Yang, Heng Tao Shen
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 geometry-guided purification to aggregated 2D features (ablation) | GeoPurify · ScanNetV2 val | mIoU: 50.2 -> 55.1 | Improved |
| add geometry-guided purification to aggregated 2D features (ablation) | GeoPurify · ScanNetV2 val | mAcc: 68.1 -> 72.5 | Improved |
| use X-Decoder instead of LSeg as 2D backbone (ablation) | GeoPurify · ScanNetV2 val | mIoU: 51.2 -> 55.1 | Improved |
| add micro-negatives to macro-negatives in contrastive sampling (ablation) | GeoPurify · ScanNetV2 val | mIoU: 53.5 -> 55.1 | Improved |
| increase pooling iterations T from 1 to 18 (ablation) | GeoPurify · ScanNetV2 val | mIoU: 52.3 -> 55.1 | Improved |
| increase pooling iterations T from 18 to 36 (ablation) | GeoPurify · ScanNetV2 val | mIoU: 55.1 -> 55.1 | No clear effect |
| increase training subset from 10 to 20 scenes (ablation) | GeoPurify · ScanNetV2 val | mIoU: 54.7 -> 55.1 | Improved |
| increase training subset from 20 to 50 scenes (ablation) | GeoPurify · ScanNetV2 val | mIoU: 55.1 -> 55.0 | No clear effect |
| select training scenes by heuristic instead of random (ablation) | GeoPurify · ScanNetV2 val | mIoU: 54.6 -> 55.1 | No clear effect |
| select training scenes by heuristic instead of coordinate/color clustering (ablation) | GeoPurify · ScanNetV2 val | mIoU: 54.9 -> 55.1 | No clear effect |
| train competitor CUA-O3D on the same ~1.5% subset instead of full data | CUA-O3D · ScanNetV2 | mIoU: 54.1 -> 18.1 | Hurt |
| use GeoPurify instead of reimplemented CUA-O3D on the 1.5% subset | GeoPurify · Matterport3D | mIoU: 14.0 -> 40.2 | Improved |
Closest papers by meaning
- Geometry-Guided Semantic Reconstruction for 3D Instance Segmentation (NeurIPS 2026 · similarity 0.89)
- Beyond 3D VQAs: Injecting 3D Spatial Priors into Vision-Language Models for Enhanced Geometric Reasoning (CVPR 2026 · similarity 0.88)
- Segment Any 3D Object with Language (ICLR 2025 · similarity 0.88)
- Mosaic3D: Foundation Dataset and Model for Open-Vocabulary 3D Segmentation (CVPR 2025 · similarity 0.87)
- Let Language Constrain Geometry: Vision–Language Models as Semantic and Spatial Critics for 3D Generation (ICML 2026 · similarity 0.87)
- XMask3D: Cross-modal Mask Reasoning for Open Vocabulary 3D Semantic Segmentation (NeurIPS 2024 · similarity 0.87)
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.
