Research map / Image generation / Image restoration
Hourglass Attention Network for Image Inpainting
Ye Deng, Siqi Hui, Rongye Meng, Sanping Zhou, Jinjun Wang
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 hourglass structure (vs standard s=1 attention) (ablation) | HAN · Paris street view, 40-50% mask | FID: 49.33 -> 46.93 | Improved |
| use hourglass order small-to-large in encoder (vs reversed spindle) (ablation) | HAN · Paris street view, 40-50% mask | FID: 48.56 -> 46.93 | Improved |
| add hourglass structure (ablation) | HAN · Paris street view, 40-50% mask | PSNR: 25.37 -> 25.67 | Improved |
| stack more hierarchical attention stages (0 to 4) (ablation) | HAN · Paris street view, 40-50% mask | FID: 68.21 -> 46.93 | Improved |
| stack more hierarchical attention stages (0 to 4) (ablation) | HAN · Paris street view, 40-50% mask | SSIM: 0.771 -> 0.834 | Improved |
| add Laplace distance prior (ablation) | HAN · Paris street view, 40-50% mask | FID: 48.67 -> 46.93 | Improved |
| use Laplace prior instead of Gaussian prior (ablation) | HAN · Paris street view, 40-50% mask | FID: 48.26 -> 46.93 | Improved |
| use Laplace prior instead of trigonometric position encoding (ablation) | HAN · Paris street view, 40-50% mask | FID: 48.17 -> 46.93 | Improved |
| use hourglass attention instead of vanilla MHA (ablation) | HAN · Model complexity | FLOPs: 183.6G -> 137.7G | Improved |
| HAN versus strongest baselines at large masks | HAN · Places2, 40-50% mask | FID: 38.69 -> 37.63 | Improved |
| HAN versus DTS at small masks on Places2 | HAN · Places2, 10-20% mask | PSNR: 28.91 -> 28.93 | No clear effect |
Closest papers by meaning
- Image Inpainting With Learnable Bidirectional Attention Maps (ICCV 2019 · similarity 0.88)
- Coherent Semantic Attention for Image Inpainting (ICCV 2019 · similarity 0.86)
- Generative Image Inpainting With Contextual Attention (CVPR 2018 · similarity 0.86)
- Incremental Transformer Structure Enhanced Image Inpainting With Masking Positional Encoding (CVPR 2022 · similarity 0.86)
- HD-Painter: High-Resolution and Prompt-Faithful Text-Guided Image Inpainting with Diffusion Models (ICLR 2025 · similarity 0.86)
- Bridging the Divide: Reconsidering Softmax and Linear Attention (NeurIPS 2024 · similarity 0.84)
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.
