Research map / Image generation / Image restoration
Image Inpainting for Irregular Holes Using Partial Convolutions
Guilin Liu, Fitsum A. Reda, Kevin J. Shih, Ting-Chun Wang, Andrew Tao, Bryan Catanzaro
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 partial convolution with mask update instead of standard convolution (ablation) | PConv U-Net inpainting · Irregular masks, hole ratio (0.2,0.3] no border | PSNR: 24.23 -> 24.54 | Improved |
| add partial convolution instead of standard convolution (ablation) | PConv U-Net inpainting · Irregular masks, hole ratio (0.2,0.3] no border | L1 error (%): 2.20 -> 2.07 | Improved |
| add partial convolution instead of standard convolution (ablation) | PConv U-Net inpainting · Irregular masks, hole ratio (0.2,0.3] no border | SSIM: 0.769 -> 0.775 | Improved |
| add partial convolution instead of standard convolution (ablation) | PConv U-Net inpainting · Irregular masks, hole ratio (0.2,0.3] no border | Inception score (lower is better): 0.603 -> 0.446 | Improved |
| add partial convolution instead of standard convolution (ablation) | PConv U-Net inpainting · Irregular masks, hole ratio (0.5,0.6] no border | PSNR: 18.24 -> 18.21 | No clear effect |
| add partial convolution instead of standard convolution (ablation) | PConv U-Net inpainting · Irregular masks, hole ratio (0.5,0.6] no border | L1 error (%): 6.66 -> 6.45 | Improved |
| add partial convolution instead of standard convolution (ablation) | PConv U-Net inpainting · Irregular masks, hole ratio (0.01,0.1] no border | L1 error (%): 0.49 vs 0.52 (PConv 0.49 worse than PM 0.45) | Hurt |
| add style loss terms (ablation) | PConv U-Net inpainting · Qualitative inpainting | visual artifacts | Improved |
| use full style loss weight instead of a small weight (ablation) | PConv U-Net inpainting · Qualitative inpainting | visual artifacts | Improved |
| add perceptual loss (ablation) | PConv U-Net inpainting · Qualitative inpainting | visual artifacts | Improved |
| add total variation loss to reduce checkerboard artifacts (ablation) | PConv U-Net inpainting · Qualitative inpainting | visual artifacts | No clear effect |
Closest papers by meaning
- Image Inpainting With Learnable Bidirectional Attention Maps (ICCV 2019 · similarity 0.90)
- Partial Convolution for Padding, Inpainting, and Image Synthesis (TPAMI 2022 · similarity 0.89)
- Deep Learning-Based Image and Video Inpainting: A Survey (IJCV 2024 · similarity 0.89)
- Generative Image Inpainting With Contextual Attention (CVPR 2018 · similarity 0.88)
- Free-Form Image Inpainting With Gated Convolution (ICCV 2019 · similarity 0.88)
- Towards Enhanced Image Inpainting: Mitigating Unwanted Object Insertion and Preserving Color Consistency (CVPR 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.
