Research map / Point clouds / Segmentation and detection
Learning Multi-View Aggregation in the Wild for Large-Scale 3D Semantic Segmentation
Check 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 |
|---|---|---|---|
| Add DeepViewAgg image features to the 3D backbone | 3D Backbone · S3DIS 6-Fold | mIoU: 69.5 -> 74.7 | Helped |
| Add DeepViewAgg image features to the 3D backbone | 3D Backbone · KITTI-360 Test | mIoU: 53.9 -> 58.3 | Helped |
| Use image features alongside the 3D network | 3D Backbone · ScanNet Val | mIoU: 69.0 -> 71.0 | Helped |
| Use average-RGB colorization instead of learned image features (ablation) | Best Configuration · S3DIS Fold 2 | mIoU: -10.8 | Hurt |
| Remove the 3D network and use image features only (ablation) | Best Configuration · S3DIS | mIoU: -5.3; -5.4; -14.5 | Mixed |
| Reduce image resolution (ablation) | Best Configuration · S3DIS Fold 2 | mIoU: -5.9 | Hurt |
| Increase 3D resolution from 5cm to 2cm (ablation) | Best Configuration · S3DIS Fold 5 | mIoU: -0.3 | No clear effect |
| Use late fusion instead of early fusion (ablation) | DeepViewAgg · S3DIS Fold 2 | mIoU: -9.1 | Hurt |
| Use one feature group instead of multiple groups (ablation) | DeepViewAgg · S3DIS Fold 2 | mIoU: -4.8 | Hurt |
| Remove view-feature gating (ablation) | DeepViewAgg · KITTI-360 Val | mIoU: -1.1 | Hurt |
| Remove dynamic image batching (ablation) | DeepViewAgg · KITTI-360 Val | mIoU: -4.5 | Hurt |
| Use featurewise max-pooling instead of learned multi-view aggregation (ablation) | DeepViewAgg · KITTI-360 Val | mIoU: -2.9 | Hurt |
Closest papers by meaning
- Geometrically-driven Aggregation for Zero-shot 3D Point Cloud Understanding (CVPR 2024 · similarity 0.89)
- ODIN: A Single Model for 2D and 3D Segmentation (CVPR 2024 · similarity 0.87)
- AGILE3D: Attention Guided Interactive Multi-object 3D Segmentation (ICLR 2024 · similarity 0.86)
- Multi-view Aggregation Network for Dichotomous Image Segmentation (CVPR 2024 · similarity 0.86)
- SUM Parts: Benchmarking Part-Level Semantic Segmentation of Urban Meshes (CVPR 2025 · similarity 0.86)
- OA-CNNs: Omni-Adaptive Sparse CNNs for 3D Semantic Segmentation (CVPR 2024 · similarity 0.86)
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.