Research map / Image generation / Editing and style transfer
Enhanced Multi-Scale Cross-Attention for Person Image Generation
Hao Tang, Ling Shao, Nicu Sebe, 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 appearance-guided shape branch (AS) to SA branch (ablation) | XingGAN++ · Market-1501 | SSIM: 0.239 -> 0.286 | Improved |
| combine SA and AS branches (ablation) | XingGAN++ · Market-1501 | SSIM: 0.286 -> 0.310 | Improved |
| add co-attention fusion (CAF) (ablation) | XingGAN++ · Market-1501 | SSIM: 0.310 -> 0.313 | Improved |
| use densely connected co-attention fusion instead of CAF (ablation) | XingGAN++ · Market-1501 | SSIM: 0.313 -> 0.317 | Improved |
| replace single-scale blocks with multi-head attention (ablation) | XingGAN++ · Market-1501 | SSIM: 0.317 -> 0.318 | No clear effect |
| replace single-scale blocks with multi-scale cross-attention (ablation) | XingGAN++ · Market-1501 | SSIM: 0.317 -> 0.328 | Improved |
| add enhanced attention (EA) module to multi-scale cross-attention (ablation) | XingGAN++ · Market-1501 | SSIM: 0.328 -> 0.333 | Improved |
| increase Xing generator from 9 to 13 blocks (ablation) | XingGAN · Market-1501 | SSIM: 0.313 -> 0.257 | Hurt |
| increase Xing++ generator from 5 to 13 blocks (ablation) | XingGAN++ · Market-1501 | SSIM: 0.333 -> 0.309 | Hurt |
| use Xing++ generator (5 blocks) instead of Xing generator (9 blocks) (ablation) | XingGAN++ · Market-1501 | SSIM: 0.313 -> 0.333 | Improved |
| use Xing block (9) instead of self-attention generator (9) (ablation) | XingGAN · Market-1501 | SSIM: 0.308 -> 0.313 | Improved |
| use cross-attention Xing generator (5 blocks) instead of ResNet generator (5 blocks) (ablation) | XingGAN · Market-1501 | SSIM: 0.297 -> 0.313 | Improved |
Closest papers by meaning
- XingGAN for Person Image Generation (ECCV 2020 · similarity 0.92)
- InsetGAN for Full-Body Image Generation (CVPR 2022 · similarity 0.86)
- Dual Contrastive Loss and Attention for GANs (ICCV 2021 · similarity 0.86)
- Diffusion-driven GAN Inversion for Multi-Modal Face Image Generation (CVPR 2024 · similarity 0.86)
- Progressive Pose Attention Transfer for Person Image Generation (CVPR 2019 · similarity 0.86)
- CG-GAN: An Interactive Evolutionary GAN-Based Approach for Facial Composite Generation (AAAI 2020 · similarity 0.86)
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.
