Research map / Image generation / Editing and style transfer
Assessing a Single Image in Reference-Guided Image Synthesis
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use RISA instead of the strongest listed baseline | single-image RIS assessment · CelebA-HQ, AFHQ, Yosemite, Church, and Bedroom | human consistency: RISA: 70.54±1.84%, 80.36±1.44%, 64.36±1.79%, 73.96±2.83%, 83.55±3.37%; baselines: 60.09±{0.99}%, 75.41±{2.02}%, 58.55±{3 | Helped |
| Train RISA on images from different model architectures than the evaluation images | RISA · Cross-model settings on CelebA-HQ and Yosemite | cross-model human consistency: 65.32±4.26%, 70.73±1.20%, 72.75±0.59%, 68.94±1.63%, 67.45±2.66%, 63.46±2.36% | Helped |
| Replace the naïve regressor with multiple binary classifiers (ablation) | RISA · CelebA-HQ, StarGAN v2 | human consistency: 67.78±{2.84}% | Helped |
| Replace the naïve regressor with multiple binary classifiers (ablation) | RISA · CelebA-HQ, StarGAN v2-Swap AE | human consistency: 53.22±{2.98}% | Helped |
| Add contrastive loss (ablation) | RISA with multiple binary classifiers · CelebA-HQ, StarGAN v2 | human consistency: 67.78±{2.84}% -> 69.52±{1.00}% | Helped |
| Add contrastive loss (ablation) | RISA with multiple binary classifiers · CelebA-HQ, StarGAN v2-Swap AE | human consistency: 53.22±{2.98}% -> 61.36±{2.82}% | Helped |
| Add supremum loss (ablation) | RISA with contrastive loss · CelebA-HQ, StarGAN v2 | human consistency: 69.52±{1.00}% -> 70.54±{2.47}% | Helped |
| Add supremum loss (ablation) | RISA with contrastive loss · CelebA-HQ, StarGAN v2-Swap AE | human consistency: 61.36±{2.82}% -> 68.94±{1.93}% | Helped |
| Add initial-stage images to stable-stage training images (ablation) | RISA training set · CelebA-HQ, StarGAN v2 | human consistency: 61.10±{4.45}% -> 66.04±{3.45}% | Helped |
| Add initial-stage images to stable-stage training images (ablation) | RISA training set · CelebA-HQ, StarGAN v2-Swap AE | human consistency: 60.42±{2.19}% -> 63.07±{1.23}% | Helped |
| Add pixel-wise interpolation to the training set (ablation) | RISA trained with initial- and stable-stage images · CelebA-HQ, StarGAN v2 | human consistency: 66.04±{3.45}% -> 70.54±{2.47}% | Helped |
| Add pixel-wise interpolation to the training set (ablation) | RISA trained with initial- and stable-stage images · CelebA-HQ, StarGAN v2-Swap AE | human consistency: 63.07±{1.23}% -> 68.94±{1.93}% | Helped |
Closest papers by meaning
- Revisiting the Evaluation of Image Synthesis with GANs (NeurIPS 2023 · similarity 0.89)
- Conditional Image Synthesis with Auxiliary Classifier GANs (ICML 2017 · similarity 0.87)
- SDEdit: Guided Image Synthesis and Editing with Stochastic Differential Equations (ICLR 2022 · similarity 0.87)
- IMAGINE: Image Synthesis by Image-Guided Model Inversion (CVPR 2021 · similarity 0.87)
- You Only Need Adversarial Supervision for Semantic Image Synthesis (ICLR 2021 · similarity 0.87)
- Unlocking Pre-trained Image Backbones for Semantic Image Synthesis (CVPR 2024 · 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.