Research map / Label-efficient and robust learning / Domain adaptation and generalization
DADA: Depth-Aware Domain Adaptation in Semantic Segmentation
Tuan-Hung Vu, Himalaya Jain, Maxime Bucher, Matthieu Cord, Patrick Perez
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 DADA fusion (depth-weighted surprisal map) to the adversarial adaptation (ablation) | DADA (S6 to S7) · SYNTHIA->Cityscapes (16 classes) | mIoU: 41.6 -> 42.6 | Improved |
| add adversarial adaptation on the auxiliary depth space (ablation) | source-only model (S1 to S4) · SYNTHIA->Cityscapes (16 classes) | mIoU: 32.2 -> 35.7 | Improved |
| add depth feature-level fusion to depth adaptation (ablation) | depth-adaptation model (S4 to S5) · SYNTHIA->Cityscapes (16 classes) | mIoU: 35.7 -> 38.0 | Improved |
| add feature fusion with depth supervision but without depth adaptation (ablation) | AdvEnt (S2 to S3) · SYNTHIA->Cityscapes (16 classes) | mIoU: 40.8 -> 40.7 | No clear effect |
| add separate adversarial depth adaptation and feature fusion to surprisal adaptation (ablation) | AdvEnt (S2 to S6) · SYNTHIA->Cityscapes (16 classes) | mIoU: 40.8 -> 41.6 | Improved |
| add adversarial surprisal adaptation to source-only training (ablation) | source-only model (S1 to S2) · SYNTHIA->Cityscapes (16 classes) | mIoU: 32.2 -> 40.8 | Improved |
| use full DADA instead of AdvEnt (depth as privileged information) | DeepLab-V2 ResNet-101 · SYNTHIA->Cityscapes (7 classes, 320x640) | mIoU: +3.2% mIoU | Improved |
| use full DADA instead of AdvEnt | DeepLab-V2 ResNet-101 · SYNTHIA->Vistas (7 classes, 320x640) | mIoU: +2.4% mIoU | Improved |
| use DADA instead of AdvEnt on person and rider classes | DADA · SYNTHIA->Cityscapes (16 classes) | per-class IoU: person 56.4 -> 54.7; rider 23.7 -> 19.3 | Hurt |
| train DADA on a fraction of SYNTHIA (50% instead of 100%) (ablation) | DADA · SYNTHIA->Cityscapes (16 classes) | mIoU: 42.6 -> 40.9 | Hurt |
| train DADA on 10% instead of 100% of SYNTHIA (ablation) | DADA · SYNTHIA->Cityscapes (16 classes) | mIoU: 42.6 -> 32.6 | Hurt |
| add depth as privileged information via DADA instead of AdvEnt | negative transfer rate · SYNTHIA->Cityscapes (7 classes, 320x640) | negative transfer rate: 11% -> 5% | Improved |
Later papers that compared with this method
18 later accepted papers report DADA as a baseline in their results tables; 18 of them report a better number on most of the results they share with it.
- Boomda: Balanced Multi-objective Optimization for Multimodal Domain Adaptation (AAAI 2026 · better on 1 of 1 shared result)
- Uncertainty-Aware Source-Free Adaptive Image Super-Resolution with Wavelet Augmentation Transformer (CVPR 2024 · better on 5 of 5 shared results)
- Hyperbolic Active Learning for Semantic Segmentation under Domain Shift (ICML 2024 · better on 2 of 2 shared results)
- Uncertainty-Aware Alignment Network for Cross-Domain Video-Text Retrieval (NeurIPS 2023 · better on 18 of 18 shared results)
- CrossMatch: Source-Free Domain Adaptive Semantic Segmentation via Cross-Modal Consistency Training (ICCV 2023 · better on 1 of 1 shared result)
- LA3: Efficient Label-Aware AutoAugment (ECCV 2022 · better on 7 of 7 shared results)
- Unsupervised Domain Adaptation for Semantic Segmentation using Depth Distribution (NeurIPS 2022 · better on 6 of 6 shared results)
- Cross-Domain Collaborative Normalization via Structural Knowledge (AAAI 2022 · better on 1 of 1 shared result)
10 more not listed.
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.
- SPIGAN — SPIGAN: Privileged Adversarial Learning from Simulation (ICLR 2019) · better on 5 of 5 shared results
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
- Universal Domain Adaptation for Semantic Segmentation (CVPR 2025 · similarity 0.92)
- Source-Free Domain Adaptation for Semantic Segmentation (CVPR 2021 · similarity 0.92)
- HRDA: Context-Aware High-Resolution Domain-Adaptive Semantic Segmentation (ECCV 2022 · similarity 0.91)
- Unsupervised Model Adaptation for Continual Semantic Segmentation (AAAI 2021 · similarity 0.91)
- Balanced Learning for Domain Adaptive Semantic Segmentation (ICML 2025 · similarity 0.90)
- Open-Set Domain Adaptation for Semantic Segmentation (CVPR 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.
