Research map / Point clouds / Segmentation and detection
Adversarially Masking Synthetic To Mimic Real: Adaptive Noise Injection for Point Cloud Segmentation Adaptation
Read the paperCheck 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 |
|---|---|---|---|
| Use adversarially learned adaptive masks instead of source-only training | SqueezeSegV3-21 · SynLiDAR → SemKITTI | mIoU: +7.6% mIoU absolute gain | Helped |
| Use adversarially learned adaptive masks instead of source-only training | SalsaNext · SynLiDAR → nuScenes | mIoU: +5.5% mIoU | Helped |
| Use both coordinate and feature embedding branches in ASM (ablation) | ASM · SynLiDAR → SemKITTI | mIoU: 21.7 -> 22.8; 22.0 -> 22.8 | Helped |
| Use binary Gumbel-Softmax masks instead of plain soft masks (ablation) | ASM · SynLiDAR → SemKITTI | mIoU: 19.6 -> 22.8 | Helped |
| Place ASM after the first convolution block rather than at the input (ablation) | SqueezeSegV3-21 · SynLiDAR → SemKITTI | mIoU: 22.0 -> 22.8 | Helped |
| Place ASM after the first convolution block rather than in the middle of the network (ablation) | SqueezeSegV3-21 · SynLiDAR → SemKITTI | mIoU: 21.7 -> 22.8 | Helped |
| Place ASM after the first convolution block rather than at the backbone end (ablation) | SqueezeSegV3-21 · SynLiDAR → SemKITTI | mIoU: 21.6 -> 22.8 | Helped |
| Update both ASM and the backbone with the adversarial generation loss (ablation) | Adversarial Masking · SynLiDAR → SemKITTI | mIoU: 21.6 -> 22.8 | Helped |
| Use adaptive learned masks rather than random spatial dropout | SqueezeSegV3-21 · SynLiDAR → SemKITTI | mIoU: 18.0±0.2 -> 22.8±0.3 | Helped |
| Use adaptive learned masks rather than random spatial dropout | SalsaNext · SynLiDAR → nuScenes | mIoU: 14.9±0.3 -> 17.0±0.3 | Helped |
| Vary the adversarial-loss coefficient λ around 1 × 10−3 (ablation) | Adversarial Masking · SynLiDAR → SemKITTI and SynLiDAR → nuScenes | mIoU: 21.9, 22.8, 21.6; 16.6, 17.0, 16.3 | Mixed |
Closest papers by meaning
- Density-guided Translator Boosts Synthetic-to-Real Unsupervised Domain Adaptive Segmentation of 3D Point Clouds (CVPR 2024 · similarity 0.89)
- Domain Adaptation on Point Clouds via Geometry-Aware Implicits (CVPR 2022 · similarity 0.87)
- Learning From Synthetic Data: Addressing Domain Shift for Semantic Segmentation (CVPR 2018 · similarity 0.86)
- Learning Semantic Segmentation From Synthetic Data: A Geometrically Guided Input-Output Adaptation Approach (CVPR 2019 · similarity 0.86)
- Adaptive Siamese Masked Autoencoder with Global Optimization for Unsupervised Point Cloud Shape Correspondence (AAAI 2025 · similarity 0.85)
- Point Cloud Pre-training with Diffusion Models (CVPR 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.