Research map / 3D vision / Face and shape
Convolutional Occupancy Networks
Songyou Peng, Michael Niemeyer, Lars Mescheder, Marc Pollefeys, Andreas Geiger
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 |
|---|---|---|---|
| replace fully-connected ONet encoding with convolutional 3-plane features | Occupancy Networks · ShapeNet point cloud reconstruction | IoU: 0.761 -> 0.884 | Improved |
| project features to three canonical planes instead of a single plane (ablation) | Ours-2D · ShapeNet point cloud reconstruction | IoU: 0.833 -> 0.884 | Improved |
| use plane projection plus 2D/3D CNN instead of direct point feature aggregation (PointConv) (ablation) | Convolutional Occupancy Networks · ShapeNet point cloud reconstruction | IoU: 0.689 -> 0.884 | Improved |
| decompose volume into three higher-resolution planes (64^2 vs 32^3) (ablation) | Convolutional Occupancy Networks · ShapeNet point cloud reconstruction | GPU memory: 5.9G -> 2.4G | Improved |
| use three planes instead of a single plane for coarse voxel input (ablation) | Ours-2D · ShapeNet voxel super-resolution | IoU: 0.652 -> 0.752 | Improved |
| three planes instead of volume on voxel super-resolution (ablation) | Convolutional Occupancy Networks · ShapeNet voxel super-resolution | GPU memory: 10.8G -> 4.0G | Improved |
| combine multi-plane and volume features (ablation) | Ours-2D (3x128^2) · Synthetic Rooms | IoU: 0.805 -> 0.816 | Improved |
| increase volume resolution from 32^3 to 64^3 (ablation) | Ours-3D · Synthetic Rooms | IoU: 0.782 -> 0.849 | Improved |
| use bilinear instead of nearest neighbor interpolation (ablation) | Ours-2D (3x128^2) · Synthetic Rooms | IoU: 0.766 -> 0.805 | Improved |
| use three planes instead of a single higher-resolution plane at similar GPU memory (ablation) | Ours-2D · Synthetic Rooms | IoU: 0.773 -> 0.805 | Improved |
| use volumetric features instead of three-plane features on real data (ablation) | Convolutional Occupancy Networks · ScanNet v2 | F-Score: 0.776 -> 0.837 | Improved |
| increase volume resolution from 32^3 to 64^3 on real data (ablation) | Ours-3D · ScanNet v2 | F-Score: 0.837 -> 0.886 | Improved |
Closest papers by meaning
- 3D Occupancy Prediction with Low-Resolution Queries via Prototype-aware View Transformation (CVPR 2025 · similarity 0.87)
- LowRankOcc: Tensor Decomposition and Low-Rank Recovery for Vision-based 3D Semantic Occupancy Prediction (CVPR 2024 · similarity 0.87)
- Unsupervised Occupancy Learning from Sparse Point Cloud (CVPR 2024 · similarity 0.87)
- AdaOcc: Adaptive 3D Occupancy Prediction for Embodied Tasks (NeurIPS 2026 · similarity 0.87)
- OctOcc: High-Resolution 3D Occupancy Prediction with Octree (AAAI 2024 · similarity 0.86)
- Achieving Speed-Accuracy Balance in Vision-based 3D Occupancy Prediction via Geometric-Semantic Disentanglement (AAAI 2025 · 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.
