Research map / Point clouds / Point cloud classification
Mitigating Geometric Degradation in Fast DownSampling via FastAdapter for Point Cloud Segmentation
Shuofeng Sun, Haibin Yan
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 FPS with random sampling without adaptation | PointMetaBase-XL · ScanNet val | mIoU: 71.8 -> 68.9 | Hurt |
| add FastAdapter and fine-tune with random sampling | PointMetaBase-XL · ScanNet val | mIoU: 68.9 -> 71.3 | Improved |
| random sampling plus FastAdapter instead of FPS | PointMetaBase-L · ScanNet val | throughput: 7.3 -> 23 ins./sec. (+315%) | Improved |
| embed FastAdapter into models already using fast downsampling | RandLA · ScanNet val | mIoU: 69.0 -> 70.0 | Improved |
| embed FastAdapter into PTV3 with grid sampling | PTV3 · ScanNet val | mIoU: 77.5 -> 78.0 | Improved |
| full fine-tuning instead of FastAdapter (ablation) | PointMetaBase-XL RS · S3DIS Area-5 | mIoU: 71.22 -> 69.42 | Hurt |
| add geometry-aware aggregation instead of plain averaging into anchors (ablation) | FastAdapter P2A · S3DIS Area-5 | mIoU: 70.12 -> 70.48 | Improved |
| add cross-layer aggregator to geometry-aware aggregation (ablation) | FastAdapter P2A · S3DIS Area-5 | mIoU: 70.48 -> 70.85 | Improved |
| add offset embedding and geometric weight instead of adding raw anchor features (ablation) | FastAdapter A2P · S3DIS Area-5 | mIoU: 69.83 -> 71.22 | Improved |
| add feature distillation from the FPS model (lambda 10 vs 0) (ablation) | FastAdapter · S3DIS Area-5 | mIoU: 70.51 -> 71.22 | Improved |
| increase anchor points beyond 100 (ablation) | FastAdapter · S3DIS Area-5 | mIoU: 71.22 (100) vs 70.92 (250) | No clear effect |
| add FastAdapter to stabilize random sampling across seeds | PointMetaBase-XL · S3DIS Area-5 | mIoU variance: ±1.1 -> ±0.3 | Improved |
Closest papers by meaning
- Point Cloud Upsampling via Disentangled Refinement (CVPR 2021 · similarity 0.88)
- FastPoint: Accelerating 3D Point Cloud Model Inference via Sample Point Distance Prediction (ICCV 2025 · similarity 0.87)
- Arbitrary-Scale Point Cloud Upsampling by Voxel-Based Network with Latent Geometric-Consistent Learning (AAAI 2024 · similarity 0.86)
- PLATYPUS: Progressive Local Surface Estimator for Arbitrary-Scale Point Cloud Upsampling (AAAI 2025 · similarity 0.86)
- A Conditional Denoising Diffusion Probabilistic Model for Point Cloud Upsampling (CVPR 2024 · similarity 0.85)
- Learning Continuous Implicit Field with Local Distance Indicator for Arbitrary-Scale Point Cloud Upsampling (AAAI 2024 · similarity 0.85)
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.
