Research map / Image generation / Autoregressive and latent generation
Wide-Context Semantic Image Extrapolation
Yi Wang, Xin Tao, Xiaoyong Shen, Jiaya Jia
Check 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 |
|---|---|---|---|
| use SRN rather than CA [48] | image extrapolation · CelebA-HQ, CUB200, DeepFashion | pairwise realism preference: 97.54%; 96.42%; 93.68% | Improved |
| add context normalization (ablation) | SRN · CelebA-HQ | PSNR: 13.92 -> 14.01 | Improved |
| add context normalization (ablation) | SRN · CelebA-HQ | SSIM: 0.5961 -> 0.6171 | Improved |
| replace confidence-driven reconstruction loss with RSV loss (ablation) | SRN · CelebA-HQ | PSNR: 14.41 -> 15.02 | Improved |
| replace confidence-driven reconstruction loss with RSV loss (ablation) | SRN · CelebA-HQ | SSIM: 0.6229 -> 0.6452 | Improved |
| replace RSV loss with vanilla l1 loss (ablation) | SRN · CelebA-HQ | PSNR: 15.02 vs 15.06 | No clear effect |
| replace RSV loss with vanilla l1 loss (ablation) | SRN · CelebA-HQ | SSIM: 0.6452 vs 0.6478 | No clear effect |
| use sub-pixel convolution instead of deconvolution or unfold for feature expansion (ablation) | SRN · CelebA-HQ | visual detail quality | Improved |
| use context adversarial loss instead of global adversarial loss alone (ablation) | SRN · CelebA-HQ | hair detail clarity | Improved |
| use small-to-large SRN rather than large-to-large SRN-HR (ablation) | image extrapolation · CelebA-HQ | visual realism | Improved |
| use small-to-large SRN rather than large-to-large SRN-HR (ablation) | image extrapolation · CelebA-HQ | inference time: 18.15 vs 28.95 | Improved |
Closest papers by meaning
- SemanticNVS: Improving Semantic Scene Understanding in Generative Novel View Synthesis (ICML 2026 · similarity 0.88)
- SemIE: Semantically-Aware Image Extrapolation (ICCV 2021 · similarity 0.88)
- Generative Powers of Ten (CVPR 2024 · similarity 0.86)
- Compositional Scene Understanding through Inverse Generative Modeling (ICML 2025 · similarity 0.86)
- Novel View Synthesis with Pixel-Space Diffusion Models (CVPR 2025 · similarity 0.85)
- Image Regeneration: Evaluating Text-to-Image Model via Generating Identical Image with Multimodal Large Language Models (AAAI 2025 · similarity 0.85)
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.
