Research map / Image generation / Image restoration
Learning a Sketch Tensor Space for Image Inpainting of Man-Made Scenes
Chenjie Cao, Yanwei Fu
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 line segment maps from LSM-HAWP (ablation) | MST · ShanghaiTech | FID: 22.68 -> 21.16 | Improved |
| add pyramid decomposing separable blocks (ablation) | MST · ShanghaiTech | FID: 21.88 -> 21.16 | Improved |
| add efficient attention module (ablation) | MST · ShanghaiTech | FID: 21.39 -> 21.16 | Improved |
| add efficient attention module (ablation) | MST · ShanghaiTech | PSNR: 26.77 -> 26.90 | Improved |
| use gated convolution only at input and output layers instead of none (ablation) | MST · ShanghaiTech | FID: 24.84 -> 22.68 | Improved |
| use gated convolution in all layers instead of only input/output (ablation) | MST · ShanghaiTech | FID: 22.68 -> 22.62 | No clear effect |
| use gated convolution in all layers instead of only input/output (ablation) | MST · ShanghaiTech | Param: 12.1M -> 25.1M | Hurt |
| add lines to canny edges in the PSS (ablation) | MST · ShanghaiTech | FID: 21.40 -> 21.16 | Improved |
| train wireframe parser with mask augmentation (LSM-HAWP) vs HAWP | wireframe parser · ShanghaiTech masked testset | sAP (threshold 5): 35.39 -> 48.93 | Improved |
| train wireframe parser with mask augmentation on uncorrupted test set | wireframe parser · ShanghaiTech unmasked testset | sAP (threshold 5): 62.16 -> 63.20 | Improved |
| replace contextual attention with efficient attention (ablation) | MST training · ShanghaiTech (batch 16) | images/sec: 19.42 -> 36.37 | Improved |
| use LSM object removal mode (m=0) instead of standard inpainting | MST · ShanghaiTech | FID: 21.16 -> 20.05 | Improved |
Closest papers by meaning
- StructureFlow: Image Inpainting via Structure-Aware Appearance Flow (ICCV 2019 · similarity 0.88)
- Contextual-based Image Inpainting: Infer, Match, and Translate (ECCV 2018 · similarity 0.87)
- SketchEdit: Mask-Free Local Image Manipulation With Partial Sketches (CVPR 2022 · similarity 0.87)
- Deep Factorised Inverse-Sketching (ECCV 2018 · similarity 0.87)
- Reconstructing the Image Stitching Pipeline: Integrating Fusion and Rectangling into a Unified Inpainting Model (NeurIPS 2024 · similarity 0.86)
- Sketch Down the FLOPs: Towards Efficient Networks for Human Sketch (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.
