Research map / Label-efficient and robust learning / Domain adaptation and generalization
MICDrop: Masking Image and Depth Features via Complementary Dropout for Domain-Adaptive Semantic Segmentation
Linyan Yang, Lukas Hoyer, Mark Weber, Tobias Fischer, Dengxin Dai, Laura Leal-Taixé, Daniel Cremers, Marc Pollefeys, 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 depth input with fusion module (no masking) (ablation) | DAFormer · GTA->Cityscapes | mIoU: 68.3 -> 69.1 | Improved |
| add complementary masking of RGB and depth features across all levels (ablation) | DAFormer + depth · GTA->Cityscapes | mIoU: 69.1 -> 70.1 | Improved |
| use independent masking of RGB and depth instead of complementary (ablation) | DAFormer + depth · GTA->Cityscapes | mIoU: 69.1 -> 69.1 | No clear effect |
| sample complementary mask independently at each feature level (ablation) | MICDrop · GTA->Cityscapes | mIoU: 70.1 -> 69.7 | Hurt |
| use combined local self-attention and global cross-attention fusion (ablation) | MICDrop fusion module · GTA->Cityscapes | mIoU: 69.7 -> 70.1 (vs local only) | Improved |
| use global depth-guided cross-attention alone for fusion (ablation) | MICDrop fusion module · GTA->Cityscapes | mIoU: 68.3 -> 68.1 (vs no-depth baseline) | Hurt |
| use CMX cross-attention fusion instead of proposed fusion (ablation) | MICDrop · GTA->Cityscapes | mIoU: 70.1 -> 68.6 | Hurt |
| freeze the pretrained RGB encoder (ablation) | DAFormer + depth · GTA->Cityscapes | mIoU: 68.7 -> 69.1 | Improved |
| add a full unmasked forward pass alongside masked mix-batch (ablation) | MICDrop training · GTA->Cityscapes | mIoU: 67.8 -> 70.1 | Improved |
| also mask source-batch features in the loss (ablation) | MICDrop training · GTA->Cityscapes | mIoU: 70.1 -> 69.9 | Hurt |
| increase mask patch size from 16 to 64 (ablation) | MICDrop · GTA->Cityscapes | mIoU: 69.4 -> 70.1 | Improved |
| plug MICDrop into MIC_HRDA | MIC_HRDA · GTA->Cityscapes | mIoU: 75.9 -> 76.6 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- DAFormer — DAFormer: Improving Network Architectures and Training Strategies for Domain-Adaptive Semantic Segmentation (CVPR 2022) · better on 3 of 3 shared results
- HRDA — HRDA: Context-Aware High-Resolution Domain-Adaptive Semantic Segmentation (ECCV 2022) · better on 2 of 2 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
- Masked Representation Modeling for Domain-Adaptive Segmentation (CVPR 2026 · similarity 0.90)
- Universal Domain Adaptation for Semantic Segmentation (CVPR 2025 · similarity 0.88)
- MaskTwins: Dual-form Complementary Masking for Domain-Adaptive Image Segmentation (ICML 2025 · similarity 0.88)
- Open-Set Domain Adaptation for Semantic Segmentation (CVPR 2024 · similarity 0.88)
- Towards Unsupervised Domain Bridging via Image Degradation in Semantic Segmentation (NeurIPS 2025 · similarity 0.87)
- Unsupervised Semantic Segmentation Through Depth-Guided Feature Correlation and Sampling (CVPR 2024 · 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.
