Research map / Detection and segmentation / 3D detection and driving
Mask3D: Pre-Training 2D Vision Transformers by Learning Masked 3D Priors
Ji Hou, Xiaoliang Dai, Zijian He, Angela Dai, Matthias Nießner
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 |
|---|---|---|---|
| Pre-train with masked RGB and sparse depth to reconstruct dense depth | ViT-B · ScanNet 2D semantic segmentation | mIoU: 63.3 -> 66.7 | Improved |
| Pre-train with masked RGB and sparse depth to reconstruct dense depth | ViT · ScanNet 2D object detection | AP@0.5: +4.3 AP@0.5 | Improved |
| Pre-train with masked RGB and sparse depth to reconstruct dense depth | ViT · ScanNet 2D instance segmentation | AP@0.5: +3.8 AP@0.5 | Improved |
| Keep only 20.0% of RGB patches and include 20.0% of depth patches (ablation) | Mask3D pre-training · ScanNet 2D semantic segmentation | mIoU: 65.2 -> 66.7 | Improved |
| Keep 20.0% rather than 50.0% of RGB patches, with 20.0% depth patches (ablation) | Mask3D pre-training · ScanNet 2D semantic segmentation | mIoU: 65.9 -> 66.7 | Improved |
| Keep 20.0% rather than 80.0% of RGB patches, with 20.0% depth patches (ablation) | Mask3D pre-training · ScanNet 2D semantic segmentation | mIoU: 64.8 -> 66.7 | Improved |
| Use 20.0% rather than 80.0% depth patches while keeping RGB at 20.0% (ablation) | Mask3D pre-training · ScanNet 2D semantic segmentation | mIoU: 65.5 -> 66.7 | Improved |
| Add RGB reconstruction loss alongside depth reconstruction (ablation) | Mask3D pre-training · ScanNet 2D semantic segmentation | mIoU: 66.7 -> 65.6 | Hurt |
| Normalize predictions and targets in the reconstruction loss (ablation) | Mask3D pre-training · ScanNet 2D semantic segmentation | mIoU: +0.8% | Improved |
| Remove Stage-I ImageNet initialization (ablation) | Mask3D pre-training · ScanNet 2D semantic segmentation | mIoU | Hurt |
| Apply Mask3D after MAE rather than training both together (ablation) | Mask3D pre-training · ScanNet 2D semantic segmentation | mIoU: 63.3 -> 66.3 | Improved |
| Apply MAE after Mask3D rather than MAE before Mask3D (ablation) | Mask3D pre-training · ScanNet 2D semantic segmentation | mIoU: 66.3 -> 63.1 | Hurt |
Later papers that compared with this method
4 later accepted papers report Mask3D as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- SAI3D: Segment Any Instance in 3D Scenes (CVPR 2024 · better on 3 of 6 shared results)
- Edge-Aware 3D Instance Segmentation Network with Intelligent Semantic Prior (CVPR 2024 · better on 2 of 2 shared results)
- Open3DIS: Open-Vocabulary 3D Instance Segmentation with 2D Mask Guidance (CVPR 2024 · better on 0 of 1 shared result)
- Mask-Attention-Free Transformer for 3D Instance Segmentation (ICCV 2023 · better on 5 of 5 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SAM-Guided Masked Token Prediction for 3D Scene Understanding (NeurIPS 2024 · similarity 0.88)
- Any3DIS: Class-Agnostic 3D Instance Segmentation by 2D Mask Tracking (CVPR 2025 · similarity 0.88)
- Geometry-Guided Semantic Reconstruction for 3D Instance Segmentation (NeurIPS 2026 · similarity 0.88)
- XMask3D: Cross-modal Mask Reasoning for Open Vocabulary 3D Semantic Segmentation (NeurIPS 2024 · similarity 0.87)
- Segment Any 3D Object with Language (ICLR 2025 · similarity 0.87)
- Masked Scene Modeling: Narrowing the Gap Between Supervised and Self-Supervised Learning in 3D Scene Understanding (CVPR 2025 · 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.
