Research map / Detection and segmentation / Open-set and few-shot detection
Abstracting Sketches through Simple Primitives
Stephan Alaniz, Massimiliano Mancini, Anjan Dutta, Diego Marcos, Zeynep Akata
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 |
|---|---|---|---|
| ground strokes to shape primitives (PMN) instead of Least Squares lines and arcs (SW) | shape-based abstraction · Quickdraw (9 classes), 10% budget | accuracy: 51.21 -> 67.08 | Improved |
| add GDSA stroke selection on top of PMN (ablation) | PMN · Quickdraw (9 classes), 10% budget | accuracy: 67.08 -> 77.22 | Improved |
| add the circle primitive to arc-only primitive set (ablation) | PMN · Quickdraw (9 classes), 10% budget | accuracy: 45.99 -> 62.48 | Improved |
| add the square primitive to the available primitive set (ablation) | PMN · ChairV2 FG-SBIR, 10% budget | top-10 accuracy: 44.58 -> 49.22 | Improved |
| use rotate-scale-rotate affine transformation instead of rotation only (ablation) | PMN · Quickdraw (9 classes), 10% budget | accuracy: 44.05 -> 67.08 | Improved |
| use rotate-scale-rotate instead of scale-rotate transformation (ablation) | PMN · Quickdraw (9 classes), 10% budget | accuracy: 64.75 -> 67.08 | Improved |
| use shear-scale-rotate instead of rotate-scale-rotate transformation (ablation) | PMN · Quickdraw (9 classes), 30% budget | accuracy: 88.86 vs 89.15 | No clear effect |
| add compatibility function phi for primitive selection at inference (ablation) | PMN · forward pass time, V100, batch size 32 | inference time (ms): 118.82 ms -> 13.14 ms | Improved |
| add compatibility function phi to the training loss (ablation) | PMN · downstream tasks | accuracy | No clear effect |
| add positional embeddings of point order to the stroke encoder (ablation) | stroke encoder · sketch tasks | performance | No clear effect |
| add OTF finetuning of the retrieval network to PMN | PMN · ShoeV2 FG-SBIR top-1, 10% budget | top-1 accuracy: 6.76 -> 9.16 | Improved |
| replace human strokes with PMN primitive reconstruction at full budget | Quickdraw classifier · Quickdraw (9 classes), 100% budget | accuracy: 97.20 -> 91.78 | Hurt |
Closest papers by meaning
- SketchAA: Abstract Representation for Abstract Sketches (ICCV 2021 · similarity 0.88)
- SketchINR: A First Look into Sketches as Implicit Neural Representations (CVPR 2024 · similarity 0.88)
- How to Handle Sketch-Abstraction in Sketch-Based Image Retrieval? (CVPR 2024 · similarity 0.88)
- It's All About Your Sketch: Democratising Sketch Control in Diffusion Models (CVPR 2024 · similarity 0.86)
- Enhance Sketch Recognition’s Explainability via Semantic Component-Level Parsing (AAAI 2024 · similarity 0.86)
- Sketch Down the FLOPs: Towards Efficient Networks for Human Sketch (CVPR 2025 · similarity 0.86)
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.
