Research map / Point clouds / Segmentation and detection
Compositional Semantic Mix for Domain Adaptation in Point Cloud Segmentation
Cristiano Saltori, Fabio Galasso, Giuseppe Fiameni, Nicu Sebe, Fabio Poiesi, Elisa Ricci
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 source-to-target mixing branch (ablation) | CoSMix t->s only · SynLiDAR to SemanticPOSS (UDA ablation) | mIoU: 31.6 -> 35.0 | Improved |
| add mean teacher update (ablation) | CoSMix both branches · SynLiDAR to SemanticPOSS (UDA ablation) | mIoU: 35.0 -> 35.4 | Improved |
| add local augmentation h (ablation) | CoSMix (d) · SynLiDAR to SemanticPOSS (UDA ablation) | mIoU: 35.4 -> 36.8 | Improved |
| add global augmentation r (ablation) | CoSMix (d) · SynLiDAR to SemanticPOSS (UDA ablation) | mIoU: 35.4 -> 37.3 | Improved |
| add both local and global augmentations (ablation) | CoSMix (d) · SynLiDAR to SemanticPOSS (UDA ablation) | mIoU: 35.4 -> 39.0 | Improved |
| add long-tail weighted class selection f (ablation) | CoSMix (g) · SynLiDAR to SemanticPOSS (UDA ablation) | mIoU: 39.0 -> 40.4 | Improved |
| raise pseudo-label confidence threshold to 0.95 (ablation) | CoSMix-UDA · SynLiDAR to SemanticPOSS (UDA ablation) | mIoU: 40.4 -> 39.2 | Hurt |
| lower pseudo-label confidence threshold to 0.65 (ablation) | CoSMix-UDA · SynLiDAR to SemanticPOSS (UDA ablation) | mIoU: 40.4 -> 40.2 | No clear effect |
| use CoSMix double-branch mixing instead of PointCutMix (ablation) | mixing strategy · SynLiDAR to SemanticPOSS (UDA ablation) | mIoU: 31.6 -> 38.9 | Improved |
| use s->t branch alone instead of t->s alone (ablation) | CoSMix mixing · SynLiDAR to SemanticPOSS (UDA ablation) | mIoU: 32.9 -> 34.8 | Improved |
| use CoSMix SSDA mixing instead of naive use of labeled target points (ablation) | CoSMix-SSDA · SynLiDAR to SemanticPOSS (SSDA) | mIoU: 33.7 -> 41.0 | Improved |
| add few labeled target samples (SSDA over UDA) | CoSMix · SynLiDAR to SemanticKITTI | mIoU: 32.2 -> 34.3 | Improved |
Closest papers by meaning
- Construct to Associate: Cooperative Context Learning for Domain Adaptive Point Cloud Segmentation (CVPR 2024 · similarity 0.91)
- Complete & Label: A Domain Adaptation Approach to Semantic Segmentation of LiDAR Point Clouds (CVPR 2021 · similarity 0.91)
- BeyondMix: Leveraging Structural Priors and Long-Range Dependencies for Domain-Invariant LiDAR Segmentation (NeurIPS 2025 · similarity 0.90)
- PointDAN: A Multi-Scale 3D Domain Adaption Network for Point Cloud Representation (NeurIPS 2019 · similarity 0.89)
- Transfer Learning from Synthetic to Real LiDAR Point Cloud for Semantic Segmentation (AAAI 2022 · similarity 0.89)
- Density-guided Translator Boosts Synthetic-to-Real Unsupervised Domain Adaptive Segmentation of 3D Point Clouds (CVPR 2024 · similarity 0.88)
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.
