Research map / Detection and segmentation / Open-set and few-shot detection
Sketch2Saliency: Learning To Detect Salient Objects From Human Drawings
Ayan Kumar Bhunia, Subhadeep Koley, Amandeep Kumar, Aneeshan Sain, Pinaki Nath Chowdhury, Tao Xiang, 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 sketches rather than text captions as the weak label | Sketch2Saliency · ECSSD | max Fβ: 0.730 -> 0.781 | Improved |
| Use sketches rather than text captions as the weak label | Sketch2Saliency · PASCAL-S | max Fβ: 0.641 -> 0.705 | Improved |
| Add class-label supervision to Sketch in the MS-WSSD framework | Sketch2Saliency MS-WSSD · ECSSD, PASCAL-S, MSRA5K, SOC, SOD, DUT-OMRON | max Fβ: 0.843 -> 0.881; 0.761 -> 0.808; 0.867 -> 0.909; 0.612 -> 0.651; 0.763 -> 0.813; 0.673 -> 0.726 | Mixed |
| Add sketch-vector normalization by using offset coordinates (ablation) | Sketch2Saliency · ECSSD | max Fβ: 0.781 minus 0.341 | Improved |
| Use 2D neighborhood attention rather than 1D attention over feature vectors (ablation) | Sketch2Saliency · ECSSD | max Fβ: 0.763 from 0.781 | Improved |
| Add equivariance loss (ablation) | Sketch2Saliency · ECSSD | max Fβ: 0.781 minus 0.021 | Improved |
| Use 20 rather than 5 Gaussians in the GMM (ablation) | Sketch2Saliency · ECSSD | max Fβ: 0.741 -> 0.781 | Improved |
| Use 20 rather than 10 Gaussians in the GMM (ablation) | Sketch2Saliency · ECSSD | max Fβ: 0.756 -> 0.781 | Improved |
| Use 20 rather than 15 Gaussians in the GMM (ablation) | Sketch2Saliency · ECSSD | max Fβ: 0.771 -> 0.781 | Improved |
| Use 20 rather than 25 Gaussians in the GMM (ablation) | Sketch2Saliency · ECSSD | max Fβ: 0.779 -> 0.781 | Improved |
| Add multi-scale feature-map aggregation (ablation) | Sketch2Saliency · ECSSD | max Fβ: 0.781 minus 0.013 | Improved |
| Use the sketch-aware sequential design rather than simple spatial attention (ablation) | Sketch2Saliency · ECSSD | max Fβ: 0.718 vs. 0.781 | Improved |
Closest papers by meaning
- Learning Dense Correspondences between Photos and Sketches (ICML 2023 · similarity 0.89)
- What Can Human Sketches Do for Object Detection? (CVPR 2023 · similarity 0.88)
- Picture That Sketch: Photorealistic Image Generation From Abstract Sketches (CVPR 2023 · similarity 0.88)
- Learning to Sketch With Shortcut Cycle Consistency (CVPR 2018 · similarity 0.88)
- Scene Categorization From Contours: Medial Axis Based Salience Measures (CVPR 2019 · similarity 0.88)
- Open Vocabulary Semantic Scene Sketch Understanding (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.
