Research map / Image generation / Autoregressive and latent generation
CR-Fill: Generative Image Inpainting With Auxiliary Contextual Reconstruction
Yu Zeng, Zhe Lin, Huchuan Lu, Vishal M. Patel
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 CR loss (ablation) | Baseline generator · Places2 val (square holes) | PSNR: 22.92 -> 23.36 | Improved |
| add CR loss (ablation) | Baseline generator · Places2 val (square holes) | L1 error: .0275 -> .0263 | Improved |
| add CR loss (ablation) | Baseline generator · Places2 val (irregular holes) | PSNR: 27.03 -> 27.79 | Improved |
| add CR loss (ablation) | Baseline generator · Places2 val (square holes) | SSIM: .8386 -> .8462 | Improved |
| replace CA layer with CR loss (ablation) | DeepFillv2 · Places2 val (square holes) | PSNR: 22.72 -> 23.37 | Improved |
| replace CA layer with CR loss (ablation) | DeepFillv2 · Places2 val (irregular holes) | PSNR: 27.03 -> 27.77 | Improved |
| replace CA layer with CR loss (ablation) | DeepFillv2 · Places2 val (square holes) | SSIM: .8429 -> 0.8447 | Mixed |
| replace CA layer with CR loss (ablation) | DeepFillv2 · Places2 val (square holes) | L1 error: .0290 -> .0262 | Improved |
| use CR loss instead of CA layer for reference patch selection (ablation) | DeepFillv2 · Places2 val (square holes, pieced reference patches) | PSNR: 21.37 -> 21.77 | Improved |
| use CR loss instead of CA layer for reference patch selection (ablation) | DeepFillv2 · Places2 val (irregular holes, pieced reference patches) | PSNR: 24.39 -> 25.27 | Improved |
| replace CA layer with CR loss (inference time) (ablation) | DeepFillv2 architecture · GPU 1024x1024 | running time (s): 0.371 -> 0.179 | Improved |
| replace CA layer with CR loss (inference time) (ablation) | DeepFillv2 architecture · CPU 2048x2048 | running time (s): 266 -> 28.9 | Improved |
Closest papers by meaning
- Generative Image Inpainting With Contextual Attention (CVPR 2018 · similarity 0.90)
- Contextual-based Image Inpainting: Infer, Match, and Translate (ECCV 2018 · similarity 0.89)
- From Inpainting to Layer Decomposition: Repurposing Generative Inpainting Models for Image Layer Decomposition (CVPR 2026 · similarity 0.89)
- Semantic Image Inpainting With Deep Generative Models (CVPR 2017 · similarity 0.89)
- Towards Enhanced Image Inpainting: Mitigating Unwanted Object Insertion and Preserving Color Consistency (CVPR 2025 · similarity 0.86)
- Brush2Prompt: Contextual Prompt Generator for Object Inpainting (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.
