Research map / Label-efficient and robust learning / Domain adaptation and generalization
I2F: A Unified Image-to-Feature Approach for Domain Adaptive Semantic Segmentation
Haoyu Ma, Xiangru Lin, Yizhou Yu
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 global photometric alignment (GPA) (ablation) | I2F (DeeplabV3+) · GTA5->Cityscapes (ablation) | mIoU (%): 51.0 -> 58.2 | Improved |
| add global manifold alignment (GMA) (ablation) | I2F (DeeplabV3+) · GTA5->Cityscapes (ablation) | mIoU (%): 56.6 -> 58.2 | Improved |
| add global texture alignment (GTEXA) (ablation) | I2F (DeeplabV3+) · GTA5->Cityscapes (ablation) | mIoU (%): 57.5 -> 58.2 | Improved |
| add target domain consistency regularization (TCR) (ablation) | I2F (DeeplabV3+) · GTA5->Cityscapes (ablation) | mIoU (%): 53.4 -> 58.2 | Improved |
| add category-oriented triplet loss (CTL) (ablation) | I2F (DeeplabV3+) · GTA5->Cityscapes (ablation) | mIoU (%): 55.1 -> 58.2 | Improved |
| use photometric+texture image adaptation instead of frequency alignment (ablation) | I2F (DeeplabV3+) · GTA5->Cityscapes (ablation) | mIoU (%): 54.0 -> 58.2 | Improved |
| use photometric+texture image adaptation instead of BDL-GAN (ablation) | I2F (DeeplabV3+) · GTA5->Cityscapes (ablation) | mIoU (%): 55.4 -> 58.2 | Improved |
| use GMA instead of adversarial global discriminator (ablation) | I2F (DeeplabV3+) · GTA5->Cityscapes (ablation) | mIoU (%): 55.8 -> 58.2 | Improved |
| apply triplet loss without target pseudo-labels instead of with (ablation) | I2F (DeeplabV3+) · GTA5->Cityscapes (ablation) | mIoU (%): 56.1 -> 58.2 | Improved |
| use hybrid GPA scheme instead of RGB mean-variance (ablation) | GPA (image adaptation) · GTA5->Cityscapes (ablation) | mIoU (%): 42.3 -> 47.3 | Improved |
| apply full I2F pipeline instead of source-only training | DeeplabV3+ · GTA5->Cityscapes (ablation) | mIoU (%): 37.6 -> 58.2 | Improved |
| add feature-level adaptation after image-level adaptation | I2F (DeeplabV3+) · GTA5->Cityscapes (ablation) | mIoU (%): 47.3 -> 58.2 | Improved |
Closest papers by meaning
- Coarse-To-Fine Domain Adaptive Semantic Segmentation With Photometric Alignment and Category-Center Regularization (CVPR 2021 · similarity 0.95)
- Universal Domain Adaptation for Semantic Segmentation (CVPR 2025 · similarity 0.94)
- Unified Domain Adaptive Semantic Segmentation (TPAMI 2025 · similarity 0.93)
- Balanced Learning for Domain Adaptive Semantic Segmentation (ICML 2025 · similarity 0.91)
- Open-Set Domain Adaptation for Semantic Segmentation (CVPR 2024 · similarity 0.91)
- Towards Unsupervised Domain Bridging via Image Degradation in Semantic Segmentation (NeurIPS 2025 · similarity 0.90)
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.
