Research map / Image generation / Autoregressive and latent generation
Contextual-based Image Inpainting: Infer, Match, and Translate
Yuhang Song, Chao Yang, Zhe Lin, Xiaofeng Liu, Qin Huang, Hao Li, C.-C. Jay Kuo
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 patch-swap and Feature2Image refinement on top of Image2Feature output (ablation) | Image2Feature output · visual comparison, 512x512 | visual quality | Improved |
| use resize-convolution instead of deconvolution in Feature2Image decoder (ablation) | Feature2Image · visual comparison | checkerboard artifacts | Improved |
| use perceptual loss instead of only l2 loss (ablation) | Feature2Image · visual comparison | sharpness | Improved |
| extract features from relu3_1 instead of relu2_1 or relu4_1 (ablation) | VGG19 feature layer · visual comparison | visual quality | Improved |
| train Feature2Image on ground truth input instead of Image2Feature prediction (ablation) | Feature2Image · visual comparison | visual quality | Improved |
| fine-tune Feature2Image with ground truth and prediction mixtures (ablation) | Feature2Image · visual comparison | visual quality | Improved |
| iteratively apply Feature2Image multiple times (ablation) | Feature2Image · visual comparison | texture sharpness and boundary artifacts | Mixed |
| use deeper fully convolutional Image2Feature with dilated layers instead of vanilla context encoder (ablation) | Image2Feature · visual comparison | visual quality | Improved |
| use our approach instead of GLI | inpainting method · ImageNet test (200 images, 512x512) | SSIM: 0.55 -> 0.56 | Improved |
| use our approach instead of NPS | inpainting method vs NPS · ImageNet test (200 images, 512x512) | Mean l1 error: 15.61% vs NPS 15.13% | Mixed |
Closest papers by meaning
- Generative Image Inpainting With Contextual Attention (CVPR 2018 · similarity 0.93)
- Semantic Image Inpainting With Deep Generative Models (CVPR 2017 · similarity 0.93)
- High-Resolution Image Inpainting Using Multi-Scale Neural Patch Synthesis (CVPR 2017 · similarity 0.92)
- Learning Pyramid-Context Encoder Network for High-Quality Image Inpainting (CVPR 2019 · similarity 0.91)
- Coherent Semantic Attention for Image Inpainting (ICCV 2019 · similarity 0.91)
- StructureFlow: Image Inpainting via Structure-Aware Appearance Flow (ICCV 2019 · similarity 0.90)
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.
