Research map / Point clouds / Segmentation and detection
SqueezeSegV3: Spatially-Adaptive Convolution for Efficient Point-Cloud Segmentation
Chenfeng Xu, Bichen Wu, Zining Wang, Wei Zhan, Peter Vajda, Kurt Keutzer, Masayoshi Tomizuka
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 SAC-ISK to SSGV3-21 baseline (ablation) | SSGV3-21 baseline · SemanticKITTI val (64x512, SSGV3-21 ablation) | mIoU (%): 44.0 -> 46.3 | Improved |
| add SAC-SK to baseline (ablation) | SSGV3-21 baseline · SemanticKITTI val (64x512, SSGV3-21 ablation) | mIoU (%): 44.0 -> 45.4 | Improved |
| add SAC-S to baseline (ablation) | SSGV3-21 baseline · SemanticKITTI val (64x512, SSGV3-21 ablation) | mIoU (%): 44.0 -> 44.9 | Improved |
| add input-channel dimension to SAC-S attention (SAC-IS) (ablation) | SAC-S · SemanticKITTI val (64x512, SSGV3-21 ablation) | mIoU (%): 44.9 -> 44.0 | Hurt |
| add kernel dimension attention to SAC-S (SAC-SK) (ablation) | SAC-S · SemanticKITTI val (64x512, SSGV3-21 ablation) | mIoU (%): 44.9 -> 45.4 | Improved |
| swap SAC-ISK for squeeze-and-excitation (SE) (ablation) | SAC-ISK · SemanticKITTI val (64x512, SSGV3-21 ablation) | mIoU (%): 46.3 -> 44.2 | Hurt |
| swap SAC-ISK for context-aggregation module (CAM) (ablation) | SAC-ISK · SemanticKITTI val (64x512, SSGV3-21 ablation) | mIoU (%): 46.3 -> 42.1 | Hurt |
| swap SAC-SK for pixel-adaptive convolution (PAC) (ablation) | SAC-SK · SemanticKITTI val (64x512, SSGV3-21 ablation) | mIoU (%): 45.4 -> 45.2 | Hurt |
| use CBAM instead of no attention (ablation) | baseline · SemanticKITTI val (64x512, SSGV3-21 ablation) | mIoU (%): 44.0 -> 44.8 | Improved |
| increase SAC attention kernel from 1x1 to 7x7 (ablation) | SAC-ISK · SemanticKITTI val (64x512, SSGV3-21 ablation) | mIoU (%): 45.5 -> 46.3 | Improved |
| increase SAC attention kernel from 3x3 to 5x5 (ablation) | SAC-ISK · SemanticKITTI val (64x512, SSGV3-21 ablation) | mIoU (%): 44.5 -> 45.4 | Mixed |
| remove last two downsampling layers and reduce channels of last two stages to 256 (ablation) | RangeNet21 · SemanticKITTI val (64x512, SSGV3-21 ablation) | mIoU (%): 38.6 -> 42.5 | Improved |
Closest papers by meaning
- Cylindrical and Asymmetrical 3D Convolution Networks for LiDAR-Based Perception (TPAMI 2021 · similarity 0.91)
- Cylindrical and Asymmetrical 3D Convolution Networks for LiDAR Segmentation (CVPR 2021 · similarity 0.91)
- Less Is More: Reducing Task and Model Complexity for 3D Point Cloud Semantic Segmentation (CVPR 2023 · similarity 0.90)
- RAPiD-Seg: Range-Aware Pointwise Distance Distribution Networks for 3D LiDAR Segmentation (ECCV 2024 · similarity 0.90)
- Large-Scale Point Cloud Semantic Segmentation With Superpoint Graphs (CVPR 2018 · similarity 0.90)
- RPVNet: A Deep and Efficient Range-Point-Voxel Fusion Network for LiDAR Point Cloud Segmentation (ICCV 2021 · similarity 0.90)
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.
