Research Paper#3D Instance Segmentation, Contrastive Learning, Computer Vision🔬 ResearchAnalyzed: Jan 3, 2026 17:08
Unified 3D Instance Segmentation with Contrastive Learning
Published:Dec 31, 2025 10:20
•1 min read
•ArXiv
Analysis
This paper addresses the challenge of inconsistent 2D instance labels across views in 3D instance segmentation, a problem that arises when extending 2D segmentation to 3D using techniques like 3D Gaussian Splatting and NeRF. The authors propose a unified framework, UniC-Lift, that merges contrastive learning and label consistency steps, improving efficiency and performance. They introduce a learnable feature embedding for segmentation in Gaussian primitives and a novel 'Embedding-to-Label' process. Furthermore, they address object boundary artifacts by incorporating hard-mining techniques, stabilized by a linear layer. The paper's significance lies in its unified approach, improved performance on benchmark datasets, and the novel solutions to boundary artifacts.
Key Takeaways
- •Proposes UniC-Lift, a unified framework for 3D instance segmentation.
- •Merges contrastive learning and label consistency steps for improved efficiency and performance.
- •Introduces a learnable feature embedding and 'Embedding-to-Label' process.
- •Addresses object boundary artifacts with hard-mining and a stabilizing linear layer.
- •Outperforms baselines on ScanNet, Replica3D, and Messy-Rooms datasets.
Reference
“The paper introduces a learnable feature embedding for segmentation in Gaussian primitives and a novel 'Embedding-to-Label' process.”