Research map / Point clouds / Segmentation and detection
Interpolation-Aware Padding for 3D Sparse Convolutional Neural Networks
Yu-Qi Yang, Peng-Shuai Wang, Yang Liu
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 |
|---|---|---|---|
| switch nearest to trilinear interpolation (zero padding) (ablation) | sparse U-Net · PartNet fine-grained (4 categories), output 64^3 | mIoU: 40.5 -> 41.5 | Improved |
| add interpolation-aware padding with trilinear interpolation (ablation) | sparse U-Net · PartNet fine-grained (4 categories), output 64^3 | mIoU: 41.5 -> 42.3 | Improved |
| use octree padding instead of zero padding (trilinear) (ablation) | sparse U-Net · PartNet fine-grained (4 categories), output 64^3 | mIoU: 41.5 -> 41.4 | No clear effect |
| use 1-ring padding instead of interpolation-aware padding (trilinear) (ablation) | sparse U-Net · PartNet fine-grained (4 categories), output 64^3 | mIoU: 42.3 -> 42.7 | Improved |
| add interpolation-aware padding without interpolation (nearest) (ablation) | sparse U-Net · PartNet fine-grained (4 categories), output 64^3 | mIoU: 40.5 -> 40.6 | Improved |
| add 1-ring padding without interpolation (nearest) (ablation) | sparse U-Net · PartNet fine-grained (4 categories), output 64^3 | mIoU: 40.5 -> 40.9 | Improved |
| add octree padding without interpolation (nearest) (ablation) | sparse U-Net · PartNet fine-grained (4 categories), output 64^3 | mIoU: 40.5 -> 40.0 | Hurt |
| coarser output resolution 32^3 with Interp+Linear vs Zero+Near (ablation) | sparse U-Net · PartNet fine-grained (4 categories), output 32^3 | mIoU: 38.1 -> 40.1 | Improved |
| apply interpolation-aware padding at all resolution levels (ablation) | sparse U-Net · PartNet (4 categories), state-of-the-art | mIoU: 42.3 -> 42.7 | Improved |
| use interpolation-aware padding and trilinear interpolation | MinkNet · ScanNet val semantic segmentation (S_out 8cm) | mIoU: 70.4 -> 72.4 | Improved |
| use interpolation-aware padding and trilinear interpolation | VoteNet with MinkNet backbone · ScanNet val detection | mAP@0.5: 37.9 -> 41.4 | Improved |
| replace PointNet++ backbone with sparse U-Net | VoteNet · ScanNet val detection | mAP@0.5: 34.7 -> 37.9 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- PointCNN — PointCNN: Convolution On X-Transformed Points (NeurIPS 2018) · better on 1 of 1 shared result
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
- Inference, Learning and Attention Mechanisms that Exploit and Preserve Sparsity in CNNs (IJCV 2020 · similarity 0.86)
- Text-guided Sparse Voxel Pruning for Efficient 3D Visual Grounding (CVPR 2025 · similarity 0.86)
- INS-Conv: Incremental Sparse Convolution for Online 3D Segmentation (CVPR 2022 · similarity 0.85)
- OA-CNNs: Omni-Adaptive Sparse CNNs for 3D Semantic Segmentation (CVPR 2024 · similarity 0.85)
- Efficient 3D Recognition with Event-driven Spike Sparse Convolution (AAAI 2025 · similarity 0.84)
- Sparse3D: Distilling Multiview-Consistent Diffusion for Object Reconstruction from Sparse Views (AAAI 2024 · similarity 0.84)
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.
