Research map / Architectures and efficiency / Attention and transformers
Attention-Based Query Expansion Learning
Albert Gordo, Filip Radenovic, Tamara Berg
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 self-attention encoders instead of fully-connected layers (ablation) | LAttQE · ROxford/RParis M/H mean | mean mAP: 66.0 -> 70.0 | Improved |
| add self-attention instead of fully-connected layers (ablation) | LAttQE · ROxford (Hard) | mAP: 41.5 -> 49.6 | Improved |
| add positional encoding (ablation) | LAttQE · ROxford (Hard) | mAP: 33.2 -> 49.6 | Improved |
| add positional encoding (ablation) | LAttQE · RParis (Hard) | mAP: 73.4 -> 70.6 | Hurt |
| add positional encoding (ablation) | LAttQE · ROxford/RParis M/H mean | mean mAP: 63.2 -> 70.0 | Improved |
| add image embedding to the weighting (not position-only) (ablation) | LAttQE · ROxford (Hard) | mAP: 42.9 -> 49.6 | Improved |
| add image embedding to the weighting (not position-only) (ablation) | LAttQE · RParis (Hard) | mAP: 66.7 -> 70.6 | Improved |
| add auxiliary classification loss (ablation) | LAttQE · ROxford/RParis M/H mean | mean mAP: 68.5 -> 70.0 | Improved |
| add auxiliary classification loss (ablation) | LAttQE · ROxford (Hard) | mAP: 47.0 -> 49.6 | Improved |
| apply LAttQE query expansion vs none (ablation) | LAttQE · ROxford (Hard) | mAP: 44.3 -> 49.6 | Improved |
| add database-side augmentation LAttDBA to LAttQE | LAttQE · ROxford/RParis (+R1M) mean | mAP (mean): 59.8 -> 63.1 | Improved |
| LAttQE vs alphaQE on ROxford Hard | query expansion · ROxford (Hard) | mAP: 44.5 -> 49.6 | Improved |
Closest papers by meaning
- QuARI: Query Adaptive Retrieval Improvement (NeurIPS 2025 · similarity 0.86)
- ARTEMIS: Attention-based Retrieval with Text-Explicit Matching and Implicit Similarity (ICLR 2022 · similarity 0.86)
- Granularity-Aware Adaptation for Image Retrieval over Multiple Tasks (ECCV 2022 · similarity 0.85)
- AMC: Attention guided Multi-modal Correlation Learning for Image Search (CVPR 2017 · similarity 0.84)
- Highlighting What Matters: Promptable Embeddings for Attribute-Focused Image Retrieval (NeurIPS 2025 · similarity 0.84)
- PQPP: A Joint Benchmark for Text-to-Image Prompt and Query Performance Prediction (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.
