Research map / 3D vision / 3D generation
Self-Supervised Pre-Training With Masked Shape Prediction for 3D Scene Understanding
Li Jiang, Zetong Yang, Shaoshuai Shi, Vladislav Golyanik, Dengxin Dai, Bernt Schiele
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 |
|---|---|---|---|
| Apply MSP pre-training | EQ-Net · ScanNet semantic segmentation | mIoU (%): 73.64 -> 75.57 | Improved |
| Apply MSP pre-training | EQ-Net with GroupFree L6,O256 · ScanNet v2 indoor detection | AP50: 51.1 -> 53.7 | Improved |
| Apply MSP pre-training | EQ-Net with CenterPoint · Waymo Level 2, 20% training data | mAPH Avg.: 63.51 -> 64.26 | Improved |
| Add shape context and deep shape feature together rather than using either alone (ablation) | MSP-SA · ScanNet semantic segmentation | mIoU (%): 75.05 / 74.92 -> 75.42 | Improved |
| Add auxiliary color prediction to the context-enhanced target (ablation) | MSP-SA · ScanNet semantic segmentation | mIoU (%): 75.42 -> 75.57 | Improved |
| Use MSP-SA with 10k sampled keypoints instead of MSP-CA or MSP-CA++ (ablation) | MSP pre-training · ScanNet semantic segmentation | mIoU (%): 74.85 / 75.05 -> 75.57 | Improved |
| Increase MSP-SA keypoint sampling from 10k to 40k (ablation) | MSP pre-training · ScanNet semantic segmentation | mIoU (%): 75.57 -> 74.49 | Hurt |
| Set masking ratio to 80% instead of 60% (ablation) | MSP-SA · ScanNet semantic segmentation | mIoU (%): 75.57 -> 74.34 | Hurt |
| Set block size to 0.5m instead of 0.3m (ablation) | MSP-SA · ScanNet semantic segmentation | mIoU (%): 75.57 -> 74.62 | Hurt |
| Replace EQ-Net with SparseConvNet as feature extractor (ablation) | MSP pre-training · ScanNet semantic segmentation | mIoU (%): 75.57 -> 74.13 | Hurt |
| Use local point set rather than context-enhanced shape target (ablation) | MSP-SA · ScanNet semantic segmentation | mIoU (%): 75.57 -> 73.71 | Hurt |
| Use color alone rather than context-enhanced shape target (ablation) | MSP-SA · ScanNet semantic segmentation | mIoU (%): 75.57 -> 73.98 | Hurt |
Closest papers by meaning
- Masked Scene Modeling: Narrowing the Gap Between Supervised and Self-Supervised Learning in 3D Scene Understanding (CVPR 2025 · similarity 0.87)
- Mask3D: Pre-Training 2D Vision Transformers by Learning Masked 3D Priors (CVPR 2023 · similarity 0.87)
- MM-3DScene: 3D Scene Understanding by Customizing Masked Modeling With Informative-Preserved Reconstruction and Self-Distilled Consistency (CVPR 2023 · similarity 0.86)
- Shape2Scene: 3D Scene Representation Learning Through Pre-training on Shape Data (ECCV 2024 · similarity 0.86)
- SAM-Guided Masked Token Prediction for 3D Scene Understanding (NeurIPS 2024 · similarity 0.84)
- 3D Feature Prediction for Masked-AutoEncoder-Based Point Cloud Pretraining (ICLR 2024 · similarity 0.83)
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.
