Research map / Label-efficient and robust learning / Recognition and long-tail
Picasso, Matisse, or a Fake? Automated Analysis of Drawings at the Stroke Level for Attribution and Authentication
Ahmed Elgammal, Yan Kang, Milko Den Leeuw
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 |
|---|---|---|---|
| combine hand-crafted stroke descriptors with GRU features | one-vs-all stroke classification · Picasso, Matisse, and Schiele across techniques | stroke test accuracy: 67.26% (8.37 %) -> 78.54% (4.36 %); 60.61% ( 8.71%) -> 77.08% ( 4.33%); 81.49% (3.30 %) -> 86.20 % (3.78%) | Improved |
| use hand-crafted features rather than GRU features for fake detection | fake drawing detection · Picasso fakes, across techniques | accuracy: 87.50 % -> 100% | Improved |
| use GRU rather than hand-crafted features | fake drawing detection · Matisse fakes, across techniques | accuracy: 76.92 % -> 100% | Improved |
| train and evaluate across techniques instead of technique-specific | GRU fake detection · Picasso fakes | accuracy: 12.50% -> 87.50 % | Improved |
| use adaptive-radius rather than fixed-radius patches (ablation) | GRU drawing classifier · across-technique classification | drawing accuracy: Picasso majority: 76.77% -> 78.11%; Matisse majority: 81.14% -> 78.79%; Schiele majority: 82.49% -> 73.40% | Mixed |
| use adaptive-radius rather than fixed-radius patches (ablation) | GRU drawing classifier · technique-specific Schiele-vs-All | drawing accuracy: 78.12% -> 90.62% | Improved |
| add Stroke Length to Stroke Thickness Profile and Curvature (ablation) | hand-crafted stroke classifier · Picasso-vs-Matisse ink strokes | test accuracy: 68.93% -> 70.17% | Improved |
| add Curvature to Fourier Descriptors, Reconstruction Error Profile, and Stroke Thickness Profile (ablation) | hand-crafted stroke classifier · Picasso-vs-Matisse ink strokes | test accuracy: 68.35% -> 74.13% | Improved |
| add Stroke Length to the full hand-crafted descriptor set (ablation) | hand-crafted stroke classifier · Picasso-vs-Matisse ink strokes | test accuracy: 74.13% -> 75.07% | Improved |
| combine hand-crafted feature groups rather than use individual descriptors (ablation) | hand-crafted stroke classifier · Picasso-vs-Matisse ink strokes | test accuracy: Fourier Descriptors 57.55%; Reconstruction Error Profile 63.20%; Stroke Thickness Profile 63.68%; Curvature 60.17%; Stro | Improved |
| use GRU instead of hand-crafted SVM features | five-class stroke classifier · Picasso, Matisse, Schiele, Modigliani, Others | test accuracy: 48.97% (5.82%) -> 74.65% (3.41%) | Improved |
Closest papers by meaning
- Neural Strokes: Stylized Line Drawing of 3D Shapes (ICCV 2021 · similarity 0.83)
- CoSE: Compositional Stroke Embeddings (NeurIPS 2020 · similarity 0.83)
- Learning Deep Sketch Abstraction (CVPR 2018 · similarity 0.83)
- What Sketch Explainability Really Means for Downstream Tasks? (CVPR 2024 · similarity 0.81)
- APDDv2: Aesthetics of Paintings and Drawings Dataset with Artist Labeled Scores and Comments (NeurIPS 2024 · similarity 0.81)
- Latent Representation Matters: Human-like Sketches in One-shot Drawing Tasks (NeurIPS 2024 · similarity 0.81)
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.
