Research map / Label-efficient and robust learning / Domain adaptation and generalization
Cross-Modal Learning for Domain Adaptation in 3D Semantic Segmentation
Maximilian Jaritz, Tuan-Hung Vu, Raoul de Charette, Émilie Wirbel, Patrick Pérez
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 cross-modal loss (xMUDA) without pseudo-labels | source-only baseline · nuScenes-Lidarseg Day/Night UDA | mIoU (2D+3D): 63.3 -> 67.4 | Improved |
| add cross-modal loss (xMUDA) without pseudo-labels | source-only baseline · VirtualKITTI/SemanticKITTI UDA | mIoU (2D+3D): 42.2 -> 48.2 | Improved |
| add pseudo-labels on top of cross-modal learning (ablation) | xMUDA · A2D2/SemanticKITTI UDA | mIoU (2D+3D): 44.0 -> 47.5 | Improved |
| add pseudo-labels on top of cross-modal learning (ablation) | xMUDA · VirtualKITTI/SemanticKITTI UDA | mIoU (2D+3D): 48.2 -> 52.0 | Improved |
| add pseudo-labels on top of cross-modal learning (ablation) | xMUDA · nuScenes-Lidarseg Day/Night UDA | mIoU (2D+3D): 67.4 -> 64.4 | Hurt |
| use cross-modal loss instead of uni-modal PL | PL baseline · VirtualKITTI/SemanticKITTI UDA | mIoU (2D+3D): 35.6 -> 48.2 | Improved |
| apply cross-modal loss on source domain as well as target (ablation) | xMUDA · nuScenes-Lidarseg USA/Singapore UDA | mIoU (2D and 3D): +1.6 mIoU on 2D and +1.3 on 3D | Improved |
| use dual-head (mimicking head) instead of single head, lambda_t=0.1 (ablation) | xMUDA · nuScenes-Lidarseg USA/Singapore UDA | mIoU | Improved |
| add cross-modal loss to supervised training | supervised segmentation · Waymo OD KRK | mIoU (2D+3D): 59.2 -> 61.1 | Improved |
| add cross-modal loss to supervised training | supervised segmentation · nuScenes-Lidarseg Singapore | mIoU (2D+3D): 79.6 -> 79.8 | No clear effect |
| use xMUDA Fusion architecture instead of vanilla late fusion UDA | fusion model · A2D2/SemanticKITTI UDA | mIoU: 38.6 -> 42.6 | Improved |
| add cross-modal learning to semi-supervised adaptation with pseudo-labels | baseline S+T_l · A2D2/SemanticKITTI SSDA | mIoU (2D+3D): 66.2 -> 70.7 | Improved |
Closest papers by meaning
- Cross-Modal Contrastive Learning for Domain Adaptation in 3D Semantic Segmentation (AAAI 2023 · similarity 0.96)
- Sparse-to-Dense Feature Matching: Intra and Inter Domain Cross-Modal Learning in Domain Adaptation for 3D Semantic Segmentation (ICCV 2021 · similarity 0.95)
- UniDxMD: Towards Unified Representation for Cross-Modal Unsupervised Domain Adaptation in 3D Semantic Segmentation (ICCV 2025 · similarity 0.92)
- Mx2M: Masked Cross-Modality Modeling in Domain Adaptation for 3D Semantic Segmentation (AAAI 2023 · similarity 0.92)
- UniDSeg: Unified Cross-Domain 3D Semantic Segmentation via Visual Foundation Models Prior (NeurIPS 2024 · similarity 0.90)
- Unsupervised Modality Adaptation with Text-to-Image Diffusion Models for Semantic Segmentation (NeurIPS 2024 · similarity 0.89)
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.
