Research map / Architectures and efficiency / Attention and transformers
Sketchformer: Transformer-Based Representation for Sketched Structure
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 dictionary-learned tokenization rather than the strongest LSTM baseline | Sketchformer on QuickDraw! · QD-862k | accuracy: 72.93 -> 78.34 | Helped |
| shuffle stroke order (ablation) | Sketchformer · QuickDraw! (QD-862k) | accuracy: 76.95 vs 77.68; 76.22 vs 77.36; 76.66 vs 78.34 | No clear effect |
| use dictionary tokenization instead of continuous stroke-5 input (ablation) | Sketchformer · QD345-L | reconstruction preference: 0.00 -> 51.4 | Helped |
| use dictionary tokenization instead of grid tokenization (ablation) | Sketchformer · QD345-L | reconstruction preference: 5.56 -> 51.4 | Helped |
| use dictionary tokenization instead of continuous input (ablation) | Sketchformer · QD345-S | reconstruction preference: 62.1 -> 24.3 | Hurt |
| use dictionary-learned tokenization (ablation) | Sketchformer · QD345-L intra-class interpolation | interpolation preference: 25.0 -> 40.6 | Helped |
| use dictionary-learned tokenization (ablation) | Sketchformer · QD345-L inter-class interpolation | interpolation preference: 21.1 -> 50.5 | Helped |
| use dictionary tokenization instead of grid tokenization (ablation) | Sketchformer SBIR · Stock10M | mAP%: 53.75 -> 56.96 | Helped |
| replace the LiveSketch sequence embedding with continuous Transformer embedding | Sketch2Sketch category retrieval · V-R | mAP%: 20.49 -> 22.22 | Helped |
| replace the LiveSketch sequence embedding with Transformer embedding | Sketch2Sketch category retrieval · R-V | mAP%: 20.93 -> 23.48 | Helped |
Closest papers by meaning
- SketchINR: A First Look into Sketches as Implicit Neural Representations (CVPR 2024 · similarity 0.91)
- How to Handle Sketch-Abstraction in Sketch-Based Image Retrieval? (CVPR 2024 · similarity 0.89)
- SketchFusion: Learning Universal Sketch Features through Fusing Foundation Models (CVPR 2025 · similarity 0.89)
- Sketch Down the FLOPs: Towards Efficient Networks for Human Sketch (CVPR 2025 · similarity 0.88)
- Enhance Sketch Recognition’s Explainability via Semantic Component-Level Parsing (AAAI 2024 · similarity 0.88)
- It's All About Your Sketch: Democratising Sketch Control in Diffusion Models (CVPR 2024 · similarity 0.88)
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.