Research map / Label-efficient and robust learning / Recognition and long-tail
Finding Badly Drawn Bunnies
Lan Yang, Kaiyue Pang, Honggang Zhang, Yi-Zhe Song
Check 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 GACL-based quality-aware recognition instead of conventional BiLSTM recognition | BiLSTM sketch recognition · QuickDraw test | top-1 accuracy: 79.87% -> 82.52% | Improved |
| Use cosine GACL rather than SketchAA | Sketch recognition · QuickDraw test | top-1 accuracy: 81.51% -> 82.52% | Improved |
| Use scale GACL rather than multiplicative angular GACL (ablation) | GACL sketch recognition · QuickDraw test | top-1 accuracy: 82.02% -> 81.77% | Hurt |
| Use additive angular GACL rather than multiplicative angular GACL (ablation) | GACL sketch recognition · QuickDraw test | top-1 accuracy: 82.02% -> 81.97% | Hurt |
| Use cosine GACL rather than multiplicative angular GACL (ablation) | GACL sketch recognition · QuickDraw test | top-1 accuracy: 82.02% -> 82.52% | Improved |
| Use cosine GACL rather than scale GACL (ablation) | GACL sketch recognition · QuickDraw test | top-1 accuracy: 81.77% -> 82.52% | Improved |
| Use cosine GACL rather than additive angular GACL (ablation) | GACL sketch recognition · QuickDraw test | top-1 accuracy: 81.97% -> 82.52% | Improved |
| Rank sketch sets using GACL feature-magnitude quality scores | GACL · QuickDraw sketch quality study, 8 categories | human agreement: 92.61% | Improved |
| Use raw feature-magnitude quality score for sketch attribution | GACL attribution · QuickDraw malicious sketch attribution | F1 score: 61.25% | Improved |
| Add stroke-removal conditioning and a qmax threshold to raw quality-score attribution (ablation) | GACL sketch attribution · QuickDraw malicious sketch attribution | F1 score | Improved |
| Lower α to prioritize quality improvement over self-reconstruction (ablation) | Quality-guided sketch generation · QuickDraw sketch generation examples | quality score | Mixed |
Closest papers by meaning
- Annotation-Free Human Sketch Quality Assessment (IJCV 2024 · similarity 0.93)
- Do Generalised Classifiers really work on Human Drawn Sketches? (ECCV 2024 · similarity 0.85)
- Learning Geometry-Aware Representations by Sketching (CVPR 2023 · similarity 0.85)
- Scoot: A Perceptual Metric for Facial Sketches (ICCV 2019 · similarity 0.84)
- SketchFusion: Learning Universal Sketch Features through Fusing Foundation Models (CVPR 2025 · similarity 0.82)
- How to Handle Sketch-Abstraction in Sketch-Based Image Retrieval? (CVPR 2024 · similarity 0.82)
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.
