Research map / Vision-language and multimodal / Human and embodied modeling
Measuring and Predicting Tag Importance for Image Retrieval
Shangwen Li, Sanjay Purushotham, Chen Chen, Yuzhuo Ren, C.‐C. Jay Kuo
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 structured model instead of Berg et al. for binary tag importance | tag importance prediction · UIUC | accuracy: 82.0% -> 87.6% | Improved |
| use continuous-valued instead of binary-valued relevance (subjective test) | retrieval relevance function · COCO Scene | human preference: 1176 out of 1500 or 78.4% | Improved |
| add visual, semantic and context features to equal-importance baseline (ablation) | structured prediction model · COCO, COCO Scene | prediction error: approximately 40% gain | Improved |
| add visual features (ablation) | tag importance model · COCO, COCO Scene | prediction error: approximately 28% reduction | Improved |
| add semantic features (ablation) | tag importance model · COCO, COCO Scene | prediction error: approximately 11% reduction | Improved |
| add context features (ablation) | tag importance model · COCO, COCO Scene | prediction error: approximately 11% reduction | Improved |
| use ground truth bounding boxes instead of detected (ablation) | tag importance model · COCO | prediction error: approximately 3% performance loss from detection error | Improved |
| add visual features alone (ablation) | tag importance model · UIUC | prediction error: 1% performance loss vs baseline | Hurt |
| use continuous instead of binary predicted tag importance | tag importance model · COCO and COCO Scene | prediction error: 48% and 45% loss for binary | Improved |
| use MIR/PCTI instead of visual baseline | image-to-image retrieval, K=50 · COCO and COCO Scene | NDCG: approximately 14% and 10% gain | Improved |
| use predicted continuous importance instead of binary-valued | MIR/PCTI vs MIR/PBTI, T2I K=50 · COCO and COCO Scene | NDCG: 2% and 5% | Improved |
| use MIR/PCTI instead of Traditional MIR | T2I K=50 · COCO and COCO Scene | NDCG: 4% and 11% | Improved |
Closest papers by meaning
- Beyond Instance-Level Image Retrieval: Leveraging Captions to Learn a Global Visual Representation for Semantic Retrieval (CVPR 2017 · similarity 0.84)
- Tagging before Alignment: Integrating Multi-Modal Tags for Video-Text Retrieval (AAAI 2023 · similarity 0.84)
- Fast Zero-Shot Image Tagging (CVPR 2016 · similarity 0.84)
- Identification of Necessary Semantic Undertakers in the Causal View for Image-Text Matching (AAAI 2024 · similarity 0.83)
- Highlighting What Matters: Promptable Embeddings for Attribute-Focused Image Retrieval (NeurIPS 2025 · similarity 0.83)
- Instance-Level Composed Image Retrieval (NeurIPS 2025 · similarity 0.82)
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.
