Research map / Label-efficient and robust learning / Domain adaptation and generalization
Generalize Then Adapt: Source-Free Domain Adaptive Semantic Segmentation
Jogendra Nath Kundu, Akshay Kulkarni, Amit Singh, Varun Jampani, R. Venkatesh Babu
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-source ERM training on augmentation groups (ablation) | single-source self-training baseline · GTA5->Cityscapes (DeepLabv2 R101) | mIoU: 44.4 -> 47.6 | Improved |
| replace ERM with Domain-experts++ (DE++) heads (ablation) | vendor-side model · GTA5->Cityscapes (DeepLabv2 R101) | mIoU: 47.6 -> 48.0 | Improved |
| replace DE++ with Leave-one-out++ (LO++) heads (ablation) | vendor-side model · GTA5->Cityscapes (DeepLabv2 R101) | mIoU: 48.0 -> 51.6 | Improved |
| add cPAE for pseudo-label extraction (ablation) | client-side self-training · GTA5->Cityscapes (DeepLabv2 R101) | mIoU: 51.6 -> 53.4 | Improved |
| add cPAE at inference (without cPAE pseudo-labels) (ablation) | client-side self-training · GTA5->Cityscapes (DeepLabv2 R101) | mIoU: 51.6 -> 52.5 | Improved |
| add cPAE at inference (with cPAE pseudo-labels) (ablation) | client-side self-training · GTA5->Cityscapes (DeepLabv2 R101) | mIoU: 53.4 -> 54.2 | Improved |
| add cPAE (pseudo-labels) on VGG-16 (ablation) | source-free adaptation · GTA5->Cityscapes (FCN8s VGG-16) | mIoU: 43.4 -> 45.9 | Improved |
| add cPAE (pseudo-labels) on SYNTHIA (ablation) | source-free adaptation · SYNTHIA->Cityscapes (16 classes, ResNet-101) | mIoU: 48.1 -> 52.0 | Improved |
| add cPAE (pseudo-labels) on SYNTHIA VGG-16 (ablation) | source-free adaptation · SYNTHIA->Cityscapes (16 classes, VGG-16) | mIoU: 40.0 -> 41.3 | Improved |
| use LO++ instead of ERM for source-only training (ablation) | vendor-side model · GTA5->Cityscapes domain generalization (ResNet-101) | mIoU: 43.1 -> 43.5 | Improved |
| use LO++ instead of ERM for source-only training (ablation) | vendor-side model · SYNTHIA->Cityscapes domain generalization (VGG16) | mIoU: 36.9 -> 37.4 | Improved |
| use the proposed augmentation groups with prior self-training method IAST | IAST · GTA5->Cityscapes (DeepLabv2 R101) | mIoU: 52.2 -> 52.8 | Improved |
Closest papers by meaning
- Unsupervised Model Adaptation for Continual Semantic Segmentation (AAAI 2021 · similarity 0.93)
- Source-Free Domain Adaptation for Semantic Segmentation (CVPR 2021 · similarity 0.92)
- Multi-Source Domain Adaptation With Collaborative Learning for Semantic Segmentation (CVPR 2021 · similarity 0.92)
- Learning to Adapt via Latent Domains for Adaptive Semantic Segmentation (NeurIPS 2021 · similarity 0.91)
- Balanced Learning for Domain Adaptive Semantic Segmentation (ICML 2025 · similarity 0.90)
- Universal Domain Adaptation for Semantic Segmentation (CVPR 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.
