Research map / Label-efficient and robust learning / Domain adaptation and generalization
QS-Attn: Query-Selected Attention for Contrastive Learning in I2I Translation
Xueqi Hu, Xinyue Zhou, Qiusheng Huang, Zhengyi Shi, Li Sun, Qingli Li
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 QS-Attn query selection and feature routing | CUT · Cityscapes | mAP: 24.7 -> 27.9 | Improved |
| add QS-Attn query selection and feature routing | CUT · Cityscapes | FID: 54.3 -> 48.8 | Improved |
| add QS-Attn query selection and feature routing | CUT · Cat→Dog | FID: 76.2 -> 72.8 | Improved |
| add QS-Attn query selection and feature routing | CUT · Horse→Zebra | SWD: 31.5 -> 30.3 | Improved |
| select queries by entropy instead of randomly (ablation) | QS-Attn global model · Horse→Zebra | FID: 52.1 -> 41.1 | Improved |
| use QS-Attn to route selected values (ablation) | QS-Attn global model · Horse→Zebra | FID: 61.1 -> 41.1 | Improved |
| route the source-domain QS-Attn matrix across both domains instead of self-domain routing (ablation) | QS-Attn global model · Horse→Zebra | FID: 43.3 -> 41.1 | Improved |
| select queries by entropy rather than randomly (ablation) | QS-Attn with self-domain routing · Horse→Zebra | FID: 53.4 -> 43.3 | Improved |
| use Informer max-mean query selection instead of entropy selection (ablation) | QS-Attn global model · Horse→Zebra | FID: 41.1 -> 43.9 | Hurt |
| use local attention for both selection and value routing instead of global attention (ablation) | QS-Attn · Horse→Zebra | FID: 41.1 -> 38.6 | Improved |
| use local selection with global value routing instead of global selection and routing (ablation) | QS-Attn · Horse→Zebra | FID: 41.1 -> 42.3 | Hurt |
| apply QS-Attn to all five feature layers instead of the last two (ablation) | QS-Attn · Cat→Dog | FID: 72.8 -> 73.1 | Hurt |
Closest papers by meaning
- Contrastive Learning for Unpaired Image-to-Image Translation (ECCV 2020 · similarity 0.87)
- TransferI2I: Transfer Learning for Image-to-Image Translation From Small Datasets (ICCV 2021 · similarity 0.85)
- Multi-Channel Attention Selection GAN With Cascaded Semantic Guidance for Cross-View Image Translation (CVPR 2019 · similarity 0.85)
- $\text{I}^2\text{AM}$: Interpreting Image-to-Image Latent Diffusion Models via Bi-Attribution Maps (ICLR 2025 · similarity 0.83)
- Rethinking the Paradigm of Content Constraints in Unpaired Image-to-Image Translation (AAAI 2024 · similarity 0.83)
- Learning Visual Generative Priors without Text (CVPR 2025 · similarity 0.83)
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.
