Research map / Point clouds / Point cloud classification
Masked Scene Contrast: A Scalable Framework for Unsupervised 3D Representation Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Replace RGB-D frame matching with scene-level point-cloud augmentation | PointContrast-style pre-training · ScanNet 20-class validation | mIoU: 74.0 -> 74.4 | Helped |
| Add masked point modeling to scene augmentation (ablation) | MSC · ScanNet 20-class validation | mIoU: 74.4 -> 75.0 | Helped |
| Increase sampled positive pairs from 1024 to 8192 (ablation) | MSC · ScanNet 20-class validation | mIoU: 74.3 -> 75.0 | Helped |
| Combine spatial and photometric augmentations (ablation) | MSC · ScanNet 20-class validation | mIoU: 72.1 -> 74.4 | Helped |
| Mix query views while leaving key views unmixed (ablation) | MSC · ScanNet 20-class validation | mIoU: 74.1 -> 74.4 | Helped |
| Mix both query and key views (ablation) | MSC · ScanNet 20-class validation | mIoU: 74.4 -> 73.7 | Hurt |
| Use cross masks together with the contrastive task (ablation) | MSC · ScanNet 20-class validation | mIoU: 74.7 -> 75.0 | Helped |
| Use a 0.1m or 0.15m mask grid rather than a 0.05m grid (ablation) | MSC · ScanNet 20-class validation | mIoU: 74.3 -> 75.0 | Helped |
| Add color reconstruction (ablation) | MSC · ScanNet 20-class validation | mIoU: 74.4 -> 74.9 | Helped |
| Add normal reconstruction (ablation) | MSC · ScanNet 20-class validation | mIoU: 74.4 -> 74.6 | Helped |
| Pre-train on scene-level point clouds rather than raw RGB-D frames | MSC · ScanNet pre-training | pre-training speedup: 4.4× | Helped |
| Pre-train MSC on ScanNet and ArkitScenes rather than ScanNet alone | MSC · ScanNet and ScanNet200 validation | mIoU | Helped |
Closest papers by meaning
- GroupContrast: Semantic-aware Self-supervised Representation Learning for 3D Understanding (CVPR 2024 · similarity 0.87)
- Masked Point-Entity Contrast for Open-Vocabulary 3D Scene Understanding (CVPR 2025 · similarity 0.87)
- Towards Compact 3D Representations via Point Feature Enhancement Masked Autoencoders (AAAI 2024 · similarity 0.87)
- Sculpting Holistic 3D Representation in Contrastive Language-Image-3D Pre-training (CVPR 2024 · similarity 0.87)
- Building a Strong Pre-Training Baseline for Universal 3D Large-Scale Perception (CVPR 2024 · similarity 0.87)
- Coherent 3D Scene Diffusion From a Single RGB Image (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.