Research map / Label-efficient and robust learning / Domain adaptation and generalization
CrossMatch: Source-Free Domain Adaptive Semantic Segmentation via Cross-Modal Consistency Training
Yifang Yin, Wenmiao Hu, Zhenguang Liu, Guanfeng Wang, Shili Xiang, Roger Zimmermann
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 cross-modal consistency regularization (ablation) | CrossMatch without denoising · GTA5->Cityscapes (19 classes) | mIoU: 50.9 -> 54.2 | Improved |
| add pseudo label denoising (ablation) | CrossMatch · GTA5->Cityscapes (19 classes) | mIoU: 50.9 -> 51.6 | Improved |
| add consistency and denoising together (ablation) | CrossMatch · GTA5->Cityscapes (19 classes) | mIoU: 50.9 -> 56.4 | Improved |
| add depth modality to the auxiliary network (ablation) | CrossMatch · GTA5->Cityscapes (19 classes) | mIoU: 50.5 -> 56.4 | Improved |
| add self-distillation stage 2 | CrossMatch · GTA5->Cityscapes (19 classes) | mIoU: 56.4 -> 57.7 | Improved |
| use multimodal auxiliary network instead of depth estimation as auxiliary task | DADA*/CorDA* · GTA5->Cityscapes (19 classes) | mIoU: 50.5 -> 56.4 | Improved |
| use multimodal auxiliary network instead of MKE | MKE* · GTA5->Cityscapes (19 classes) | mIoU: 51.5 -> 56.4 | Improved |
| replace DeepLabv2 target model with SegFormer in plain self-training | self-training · GTA5->Cityscapes (19 classes) | mIoU: 50.5 -> 51.3 | Improved |
| apply CrossMatch on top of GtA w/ cPAE | GtA w/ cPAE · GTA5->Cityscapes (19 classes) | mIoU: 53.4 -> 57.3 | Improved |
| apply CrossMatch on top of ProDA | ProDA · GTA5->Cityscapes (19 classes) | mIoU: 57.5 -> 59.5 | Improved |
| vary balancing coefficient gamma from 0.5 to 10 (ablation) | CrossMatch · GTA5->Cityscapes (19 classes) | mIoU: 55.7 to 56.7 | No clear effect |
| increase SSA-Gates in auxiliary network from 1 to 4 (ablation) | auxiliary network · GTA5->Cityscapes (19 classes) | mIoU: 43.4 -> 56.6 | 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.
- DADA* — DADA: Depth-Aware Domain Adaptation in Semantic Segmentation (ICCV 2019) · 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
- Cross-Modal Learning for Domain Adaptation in 3D Semantic Segmentation (TPAMI 2022 · similarity 0.92)
- Generalize Then Adapt: Source-Free Domain Adaptive Semantic Segmentation (ICCV 2021 · similarity 0.91)
- Sparse-to-Dense Feature Matching: Intra and Inter Domain Cross-Modal Learning in Domain Adaptation for 3D Semantic Segmentation (ICCV 2021 · similarity 0.91)
- Bi-directional Contrastive Learning for Domain Adaptive Semantic Segmentation (ECCV 2022 · similarity 0.90)
- Unsupervised Modality Adaptation with Text-to-Image Diffusion Models for Semantic Segmentation (NeurIPS 2024 · similarity 0.89)
- Connectivity-Driven Pseudo-Labeling Makes Stronger Cross-Domain Segmenters (NeurIPS 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.
