Research map / Image generation / Autoregressive and latent generation
Learning Deep Sketch Abstraction
Umar Riaz Muhammad, Yongxin Yang, Yi-Zhe Song, Tao Xiang, Timothy M. Hospedales
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 |
|---|---|---|---|
| use ranked reward instead of basic reward (ablation) | Sketch abstraction model · QuickDraw test, human sketches, abstraction levels 1–3 | accuracy: 87.60% -> 88.20%; 78.80% -> 80.80%; 68.00% -> 70.40% | Improved |
| use ranked reward instead of basic reward (ablation) | Sketch abstraction model · QuickDraw test, synthesized sketches, abstraction levels 1–3 | accuracy: 92.60% -> 94.20%; 88.40% -> 90.80%; 73.20% -> 79.40% | Improved |
| learn stroke-segment removal rather than remove segments randomly (ablation) | Sketch abstraction model · QuickDraw test, human sketches, abstraction levels 1–3 | accuracy: 85.00% -> 88.20%; 74.60% -> 80.80%; 64.20% -> 70.40% | Improved |
| learn stroke-segment removal rather than remove segments randomly (ablation) | Sketch abstraction model · QuickDraw test, synthesized sketches, abstraction levels 1–3 | accuracy: 89.96% -> 94.20%; 80.20% -> 90.80%; 69.20% -> 79.40% | Improved |
| train FG-SBIR with sketches synthesized by the abstraction pipeline instead of simplified edge-maps (ablation) | FG-SBIR model · Shoe-V2 | Top1 matching accuracy: 16.67% -> 21.17% | Improved |
| train FG-SBIR with sketches synthesized by the abstraction pipeline instead of simplified edge-maps (ablation) | FG-SBIR model · Chair-V2 | Top1 matching accuracy: 34.67% -> 41.80% | Improved |
| train FG-SBIR with sketches synthesized by the abstraction pipeline instead of the cited synthesis model (ablation) | FG-SBIR model · Shoe-V2 | Top1 matching accuracy: 8.86% -> 21.17% | Improved |
| train FG-SBIR with sketches synthesized by the abstraction pipeline instead of the cited synthesis model (ablation) | FG-SBIR model · Chair-V2 | Top1 matching accuracy: 31.27% -> 41.80% | Improved |
| include multiple abstraction levels during FG-SBIR training and testing | FG-SBIR model · Shoe-V2 and Chair-V2 | Top1 matching accuracy | Improved |
| replace the softmax temperature adjustment with a probability-distribution shift to control abstraction | Sketch abstraction model · Sketch abstraction | abstraction-level control | Improved |
| generate photo-to-sketch outputs by edge-map preprocessing and abstraction rather than encoder-decoder image translation | Photo-to-sketch synthesis · QMUL Shoe-V2 | visual appeal | Improved |
| use the learned abstraction model rather than random stroke removal | Sketch abstraction model · Human study | aesthetic preference: 64.3 ± 4.59 | Improved |
Closest papers by meaning
- SketchAA: Abstract Representation for Abstract Sketches (ICCV 2021 · similarity 0.90)
- How to Handle Sketch-Abstraction in Sketch-Based Image Retrieval? (CVPR 2024 · similarity 0.90)
- It's All About Your Sketch: Democratising Sketch Control in Diffusion Models (CVPR 2024 · similarity 0.89)
- SketchFusion: Learning Universal Sketch Features through Fusing Foundation Models (CVPR 2025 · similarity 0.88)
- What Sketch Explainability Really Means for Downstream Tasks? (CVPR 2024 · similarity 0.87)
- SketchINR: A First Look into Sketches as Implicit Neural Representations (CVPR 2024 · similarity 0.87)
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.
