Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Do Generalised Classifiers really work on Human Drawn Sketches?
Hmrishav Bandyopadhyay, Pinaki Nath Chowdhury, Aneeshan Sain, Subhadeep Koley, Tao Xiang, Ayan Kumar Bhunia, Yi-Zhe Song
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 sketch-specific prompt learning on CLIP (Ours) instead of MaPLe | CLIP-based sketch classifier · Edgemaps, TU-Berlin, QuickDraw few-shot | top-1 accuracy (%): average margin of 6.66% over MaPLe | Improved |
| use deep prompts (depth 9) instead of shallow prompts (depth 1) (ablation) | Ours · TU-Berlin seen classes | top-1 accuracy (%): 66.87 -> 76.96 | Improved |
| add Meta-Net for instance-conditioned prompts (ablation) | Ours · TU-Berlin unseen classes | top-1 accuracy (%): 54.19 -> 60.51 | Improved |
| add raster-to-vector auxiliary loss (ablation) | Ours · TU-Berlin seen classes | top-1 accuracy (%): 73.70 -> 76.96 | Improved |
| add abstraction codebook vectors and mixup (ablation) | Ours · TU-Berlin seen classes | top-1 accuracy (%): 73.09 -> 76.96 | Improved |
| add abstraction-mixup augmentation (ablation) | Ours · TU-Berlin seen classes | top-1 accuracy (%): 71.36 -> 76.96 | Improved |
| add abstraction mixup on unseen classes (ablation) | Ours · TU-Berlin unseen classes | top-1 accuracy (%): 60.00 -> 60.51 | No clear effect |
| use 5 context tokens instead of 20 (ablation) | Ours · TU-Berlin unseen classes | top-1 accuracy (%): 55.94 -> 60.51 | Improved |
| use CLIP-based methods instead of non-CLIP sketch baselines for unseen classes | zero-shot sketch recognition · TU-Berlin unseen classes | top-1 accuracy (%): ResNet 1.82 -> CLIP-Z 47.71 | Improved |
Closest papers by meaning
- SketchAA: Abstract Representation for Abstract Sketches (ICCV 2021 · similarity 0.90)
- SketchINR: A First Look into Sketches as Implicit Neural Representations (CVPR 2024 · similarity 0.89)
- SketchFusion: Learning Universal Sketch Features through Fusing Foundation Models (CVPR 2025 · similarity 0.89)
- 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 0.88)
- What Sketch Explainability Really Means for Downstream Tasks? (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.
