Research map / Detection and segmentation / Weakly supervised detection and segmentation
How Hard Can It Be? Estimating the Difficulty of Visual Search in an Image
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 CNN deep features with ν-SVR rather than hand-engineered image cues | Visual search difficulty prediction · PASCAL VOC 2012 test | Kendall τ: 0.299 -> 0.472 | Helped |
| Use ν-SVR instead of KRR (ablation) | VGG-f difficulty predictor · PASCAL VOC 2012 test | Kendall τ: 0.345 -> 0.440 | Helped |
| Add pyramid features (ablation) | VGG-f difficulty predictor · PASCAL VOC 2012 test | Kendall τ: 0.440 -> 0.458 | Helped |
| Add horizontal-flip features (ablation) | VGG-f difficulty predictor · PASCAL VOC 2012 test | Kendall τ: 0.458 -> 0.459 | Helped |
| Add pyramid features (ablation) | VGG-vd difficulty predictor · PASCAL VOC 2012 test | Kendall τ: 0.442 -> 0.467 | Helped |
| Add horizontal-flip features (ablation) | VGG-vd difficulty predictor · PASCAL VOC 2012 test | Kendall τ: 0.467 -> 0.468 | Helped |
| Combine VGG-f and VGG-vd features (ablation) | CNN difficulty predictor · PASCAL VOC 2012 test | Kendall τ: 0.468 -> 0.472 | Helped |
| Order training images from easiest to hardest | MIL for weakly supervised object localization · PASCAL VOC 2007 train+validation | CorLoc: 34.4% -> 42.8% | Helped |
| Select easiest unlabeled examples using predicted difficulty | Semi-supervised object classification · PASCAL VOC 2012 validation, 7 classes | mAP: 87.8% ± 0.6% -> 88.4% ± 0.6% | Helped |
| Select unlabeled examples randomly (ablation) | Semi-supervised object classification · PASCAL VOC 2012 validation, 7 classes | mAP: 87.8% ± 0.6% -> 86.1% ± 1.0% | Hurt |
| Select most confident unlabeled examples (ablation) | Semi-supervised object classification · PASCAL VOC 2012 validation, 7 classes | mAP | No clear effect |
| Select least confident unlabeled examples (ablation) | Semi-supervised object classification · PASCAL VOC 2012 validation, 7 classes | mAP: 85.2% ± 1.1% | Hurt |
Closest papers by meaning
- How hard are computer vision datasets? Calibrating dataset difficulty to viewing time (NeurIPS 2023 · similarity 0.90)
- Query-Based Hard-Image Retrieval for Object Detection at Test Time (AAAI 2023 · similarity 0.87)
- Interpreting Object-level Foundation Models via Visual Precision Search (CVPR 2025 · similarity 0.84)
- CVSearch: Empowering Multimodal LLMs with Cognitive Visual Search for High-Resolution Image Perception (ICML 2026 · similarity 0.83)
- Visual Riddles: a Commonsense and World Knowledge Challenge for Large Vision and Language Models (NeurIPS 2024 · similarity 0.83)
- Re-thinking Temporal Search for Long-Form Video Understanding (CVPR 2025 · similarity 0.83)
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.