Research map / Label-efficient and robust learning / Domain adaptation and generalization
mDALU: Multi-Source Domain Adaptation and Label Unification With Partial Datasets
Rui Gong, Dengxin Dai, Yuhua Chen, Wen Li, Luc Van Gool
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 multi-branch network (MBR) (ablation) | source-only baseline · Digits classification, non-overlapping source labels (avg of MT/SYN/SVHN targets) | accuracy (%): 60.65 -> 61.62 | Improved |
| add uncertainty maximization (UM) (ablation) | MBR · Digits classification, non-overlapping source labels (avg of MT/SYN/SVHN targets) | accuracy (%): 61.62 -> 73.08 | Improved |
| add the A3 module's domain attention (DAT, no adversarial part) (ablation) | MBR + UM · Digits classification, non-overlapping source labels (avg of MT/SYN/SVHN targets) | accuracy (%): 73.08 -> 74.90 | Improved |
| add adversarial alignment to A3 (ablation) | MBR + UM + DAT · Digits classification, non-overlapping source labels (avg of MT/SYN/SVHN targets) | accuracy (%): 74.90 -> 75.13 | Improved |
| add pseudo-label supervision fusion (PSF) (ablation) | mDALU w/o PSF · Digits classification, non-overlapping source labels (avg of MT/SYN/SVHN targets) | accuracy (%): 75.13 -> 79.00 | Improved |
| apply domain adaptation methods DANN/DCTN/M3SDA without label unification | source-only · Digits classification, non-overlapping source labels (avg of MT/SYN/SVHN targets) | accuracy (%): 60.65 -> 59.75 | Hurt |
| use ambiguity cross entropy label unification (AENT) | source-only · Digits classification, non-overlapping source labels (avg of MT/SYN/SVHN targets) | accuracy (%): 60.65 -> 64.90 | Improved |
| add PSF (ablation) | mDALU w/o PSF · Digits, 4 source domains (avg) | accuracy (%): 80.83 -> 82.88 | Improved |
| add PSF (ablation) | mDALU w/o PSF · Cross-modal 2D (Cityscapes+Nuscenes->A2D2) | mIoU (%): 38.1 -> 54.9 | Improved |
| add PSF (ablation) | mDALU w/o PSF · Cross-modal 3D (Cityscapes+Nuscenes->A2D2) | mIoU (%): 2.4 -> 37.1 | Improved |
| add inference weight map for inconsistent taxonomies (ablation) | Ours w/o PSF · GTA5+SYNTHIA->Cityscapes (T, inconsistent taxonomies) | mIoU (%): 35.3 -> 37.2 | Improved |
| add PSF with relabeling of inconsistent classes (ablation) | Ours (PSF w/o relabeling) · GTA5+SYNTHIA->Cityscapes (T, inconsistent taxonomies) | mIoU (%): 38.4 -> 40.0 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- xMUDA — xMUDA: Cross-Modal Unsupervised Domain Adaptation for 3D Semantic Segmentation (CVPR 2020) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Generalize Then Adapt: Source-Free Domain Adaptive Semantic Segmentation (ICCV 2021 · similarity 0.89)
- Semantic Concentration for Domain Adaptation (ICCV 2021 · similarity 0.89)
- Domain Adaptive Semantic Segmentation Using Weak Labels (ECCV 2020 · similarity 0.89)
- Label-Driven Reconstruction for Domain Adaptation in Semantic Segmentation (ECCV 2020 · similarity 0.88)
- A Theory of Learning Unified Model via Knowledge Integration from Label Space Varying Domains (CVPR 2025 · similarity 0.88)
- Balanced Learning for Domain Adaptive Semantic Segmentation (ICML 2025 · similarity 0.87)
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.
