Research map / Detection and segmentation / Open-set and few-shot detection
Self-Supervised Image Representation Learning With Geometric Set Consistency
Nenglun Chen, Lei Chu, Hao Pan, Yan Lu, Wenping Wang
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 geometric consistency set-InfoNCE after Pri3D View and Geo pre-training | Ours · ScanNet semantic segmentation, ResNet50 | mIoU: 61.7 -> 63.1 | Improved |
| Add geometric consistency set-InfoNCE after Pri3D View and Geo pre-training | Ours · NYUv2 semantic segmentation | mIoU: 54.7 -> 55.4 | Improved |
| Add geometric consistency set-InfoNCE after Pri3D View and Geo pre-training | Ours · ScanNet object detection | AP: 26.6 -> 26.9 | Improved |
| Add geometric consistency set-InfoNCE after Pri3D View and Geo pre-training | Ours · ScanNet instance segmentation | AP: 18.7 -> 19.5 | Improved |
| Add geometric consistency set-InfoNCE after Pri3D View and Geo pre-training | Ours · NYUv2 object detection | AP: 19.4 -> 19.7 | Improved |
| Add geometric consistency set-InfoNCE after Pri3D View and Geo pre-training | Ours · NYUv2 instance segmentation | AP: 15.8 -> 16.5 | Improved |
| Use asymmetric set feature aggregation rather than averaging all set features (ablation) | Geometric set-InfoNCE · ScanNet semantic segmentation | mIoU: 63.1 vs 63.1 | No clear effect |
| Vary geometric set clustering edge threshold from 0.01 to 0.05 (ablation) | Geometric consistency set pre-training · ScanNet semantic segmentation | mIoU: 63.2; 62.9; 62.9; 63.2; 63.1 | No clear effect |
| Replace geometry-derived consistency sets with ground-truth semantic sets (ablation) | Set-InfoNCE pre-training · ScanNet semantic segmentation | mIoU: 63.1 -> 66.4 | Improved |
| Continue Pri3D training with View loss (ablation) | Pri3D · ScanNet semantic segmentation | mIoU: 61.7 -> 61.4 | Hurt |
| Continue Pri3D training with Geo loss (ablation) | Pri3D · ScanNet semantic segmentation | mIoU: 61.7 -> 61.6 | Hurt |
| Continue Pri3D training with View and Geo losses (ablation) | Pri3D · ScanNet semantic segmentation | mIoU: 61.7 -> 61.5 | Hurt |
Closest papers by meaning
- Learning 3D Representations for Spatial Intelligence from Unposed Multi-View Images (CVPR 2026 · similarity 0.87)
- Spatially Consistent Representation Learning (CVPR 2021 · similarity 0.87)
- Geometry-Guided Semantic Reconstruction for 3D Instance Segmentation (NeurIPS 2026 · similarity 0.86)
- Improving Semantic Correspondence with Viewpoint-Guided Spherical Maps (CVPR 2024 · similarity 0.85)
- Building a Strong Pre-Training Baseline for Universal 3D Large-Scale Perception (CVPR 2024 · similarity 0.85)
- Near, far: Patch-ordering enhances vision foundation models' scene understanding (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.
