Research map / Vision-language and multimodal / Human and embodied modeling
Sketching With Style: Visual Search With Sketches and Aesthetic Context
John Collomosse, Tu Bui, Michael J. Wilber, Chen Fang, Hailin Jin
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 |
|---|---|---|---|
| add triplet refinement after softmax training of the style model (ablation) | style-softmax · Behance-Net-TT style-only retrieval (1000 queries) | mAP: 53.6 -> 61.0 | Improved |
| train on semantically balanced instead of unbalanced triplets (ablation) | style-triplet · Behance-Net-TT style-only retrieval (1000 queries) | mAP: 55.0 -> 61.0 | Improved |
| add bottleneck to GoogLeNet features (ablation) | GoogLeNet · Behance-Net-TT style-only retrieval (1000 queries) | mAP: 46.8 -> 53.6 | Improved |
| learn style on artwork with triplet loss instead of Gram matrices | Gram / Gatys et al. · Behance-Net-TT style-only retrieval (1000 queries) | mAP: 38.4 -> 61.0 | Improved |
| use 64-D output layer instead of 128-D (ablation) | hierarchical network · Behance-VS sketch+style (MTurk, 24 sketches x 11 styles) | mAP: 48.6 -> 52.7 | Improved |
| learn joint projection of sketch and style instead of concatenation (ablation) | ss-concat · Behance-VS sketch+style (MTurk, 24 sketches x 11 styles) | mAP: 47.2 -> 52.7 | Improved |
| use proposed ss-triplet network instead of baselines | GoogLeNet · Behance-VS sketch+style (MTurk, 24 sketches x 11 styles) | mAP: 38.8 -> 52.7 | Improved |
| use visual style set instead of free text for aesthetic context (ablation) | ss-concat · Behance-VS reduced set (4.2k images, 88 queries) | mAP: 18.6 -> 40.1 | Improved |
| use sketch instead of keywords as structure query (ablation) | ss-concat · Behance-VS reduced set (4.2k images, 88 queries) | mAP: 36.4 -> 40.1 | Improved |
| compare structure relevance across methods (ablation) | all methods · Behance-VS sketch+style (MTurk, 24 sketches x 11 styles) | structure relevance: 62.5±2.3 | No clear effect |
Closest papers by meaning
- How to Handle Sketch-Abstraction in Sketch-Based Image Retrieval? (CVPR 2024 · similarity 0.90)
- ALADIN: All Layer Adaptive Instance Normalization for Fine-Grained Style Similarity (ICCV 2021 · similarity 0.90)
- StyleMeUp: Towards Style-Agnostic Sketch-Based Image Retrieval (CVPR 2021 · similarity 0.90)
- Doodle to Search: Practical Zero-Shot Sketch-Based Image Retrieval (CVPR 2019 · similarity 0.89)
- Sketch Down the FLOPs: Towards Efficient Networks for Human Sketch (CVPR 2025 · similarity 0.88)
- SketchFusion: Learning Universal Sketch Features through Fusing Foundation Models (CVPR 2025 · 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.
