Research map / 3D vision / 3D generation
SSR-2D: Semantic 3D Scene Reconstruction From 2D Images
Junwen Huang, Alexey Artemov, Yujin Chen, Shuaifeng Zhi, Kai Xu, 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 |
|---|---|---|---|
| add virtual view selection (2D GT labels) (ablation) | Ours · Matterport3D SG | mIoU (%): 26.9 -> 30.1 | Improved |
| add virtual view selection (pseudo-labels) (ablation) | Ours · Matterport3D SG | mIoU (%): 20.5 -> 21.5 | Improved |
| increase virtual views from 5 to 15 (2D GT) (ablation) | Ours · Matterport3D SG | mIoU (%): 30.1 -> 32.6 | Improved |
| add geometry completion head (ablation) | Ours · Matterport3D SG | mIoU (%): 32.8 -> 36.7 | Improved |
| add color encoder (ablation) | Ours · Matterport3D SSC | mIoU (%): 22.4 -> 28.6 | Improved |
| add color decoder (ablation) | Ours · Matterport3D SSC | mIoU (%): 26.3 -> 28.6 | Improved |
| train with both 2D and 3D ground truth instead of 2D only (ablation) | Ours · Matterport3D SSC | mIoU (%): 30.1 -> 35.1 | Improved |
| use machine-generated 2D pseudo-labels instead of 2D GT (ablation) | Ours · Matterport3D SSC | mIoU (%): 30.1 -> 21.5 | Hurt |
| increase input frame completeness from 30-40% to 40-50% (ablation) | Ours · Matterport3D SG | mIoU (%): 36.7 -> 40.4 | Improved |
| train on ScanNet and test on Matterport3D instead of in-domain (ablation) | Ours · Matterport3D SSC (cross-data) | mIoU (%): 36.7 -> 17.9 | Hurt |
| use Ours instead of Semantic-NeRF | Ours · ScanNet SSC | Geo. mIoU: 3.9 -> 49.8 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- ScanComplete — ScanComplete: Large-Scale Scene Completion and Semantic Segmentation for 3D Scans (CVPR 2018) · better on 6 of 6 shared results
- SPSG — SPSG: Self-Supervised Photometric Scene Generation From RGB-D Scans (CVPR 2021) · better on 2 of 2 shared results
- SGNN — SG-NN: Sparse Generative Neural Networks for Self-Supervised Scene Completion of RGB-D Scans (CVPR 2020) · better on 2 of 2 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
- ScanNet: Richly-Annotated 3D Reconstructions of Indoor Scenes (CVPR 2017 · similarity 0.91)
- 3DMV: Joint 3D-Multi-View Prediction for 3D Semantic Scene Segmentation (ECCV 2018 · similarity 0.90)
- Point Cloud Semantic Scene Completion from RGB-D Images (AAAI 2021 · similarity 0.90)
- 3D-to-2D Distillation for Indoor Scene Parsing (CVPR 2021 · similarity 0.90)
- Large Spatial Model: End-to-end Unposed Images to Semantic 3D (NeurIPS 2024 · similarity 0.89)
- Semi-supervised 3D Semantic Scene Completion with 2D Vision Foundation Model Guidance (AAAI 2025 · similarity 0.88)
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.
