Research map / Vision-language and multimodal / Human and embodied modeling
LiveSketch: Query Perturbations for Guided Sketch-Based Visual Search
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 vector stroke-sequence queries rather than rasterized queries (ablation) | LiveSketch search embedding · Stock67M, QD-345 | class-level mAP (%): 35.26 -> 38.40 | Helped |
| Use vector stroke-sequence queries rather than rasterized queries (ablation) | LiveSketch search embedding · Stock67M, QD-345 | instance-level mAP (%): 29.48 -> 30.81 | Helped |
| Add the joint search embedding rather than using the intermediate raster embedding (ablation) | LiveSketch · Stock67M, QD-345 | class-level mAP (%): 35.15 -> 38.40 | Helped |
| Add the joint search embedding rather than using the intermediate raster embedding (ablation) | LiveSketch · Stock67M, QD-345 | instance-level mAP (%): 27.48 -> 30.81 | Helped |
| Add interactive sketch suggestions and iterative refinement (ablation) | LiveSketch · Time-to-task, instance-level | seconds: 45.43 -> 30.74 | Helped |
| Add interactive sketch suggestions and iterative refinement (ablation) | LiveSketch · Time-to-task, mean average | seconds: 41.72 -> 27.67 | Helped |
| Add interactive sketch suggestions and iterative refinement (ablation) | LiveSketch · Time-to-task, class-level | seconds: 38.33 -> 24.90 | Helped |
| Preserve vector stroke order instead of shuffling strokes (ablation) | LiveSketch cross-modal matching · QD-173K, QD-345, instance-level | mAP (%): 15.71 -> 18.80 | Helped |
| Preserve vector stroke order instead of shuffling strokes (ablation) | LiveSketch cross-modal matching · QD-173K, QD-345, class-level | mAP (%): 35.94 -> 34.88 | Hurt |
| Replace nonlinear backpropagation interpolation with direct linear interpolation (ablation) | LiveSketch query suggestions · Inter-category interpolation | human preference (%): 18.0% vs. 26.7% | Hurt |
| Replace nonlinear backpropagation interpolation with direct linear interpolation (ablation) | LiveSketch query suggestions · Intra-category interpolation | human preference (%): 18.7% vs. 25.3 | Hurt |
| Add the joint search embedding instead of using the intermediate raster embedding (ablation) | LiveSketch time-to-task · Time-to-task, instance-level | seconds: 95.27 -> 30.74 | Hurt |
Closest papers by meaning
- Sketch Me That Shoe (CVPR 2016 · similarity 0.86)
- Sketch Down the FLOPs: Towards Efficient Networks for Human Sketch (CVPR 2025 · similarity 0.86)
- How to Handle Sketch-Abstraction in Sketch-Based Image Retrieval? (CVPR 2024 · similarity 0.85)
- SketchINR: A First Look into Sketches as Implicit Neural Representations (CVPR 2024 · similarity 0.84)
- O3SLM: Open Weight, Open Data, and Open Vocabulary Sketch-Language Model (AAAI 2026 · similarity 0.84)
- Composite Sketch+Text Queries for Retrieving Objects with Elusive Names and Complex Interactions (AAAI 2024 · similarity 0.84)
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.