Research map / Vision-language and multimodal / Human and embodied modeling
Drill-down: Interactive Retrieval of Complex Scenes using Natural Language Queries
Fuwen Tan, Paola Cascante-Bonilla, Xiaoxiao Guo, Hui Wu, Song Feng, Vicente Ordonez
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 region features instead of whole-image features (HRE -> R-HRE) (ablation) | HRE · Visual Genome simulated queries | R@K | Improved |
| hierarchical context encoding instead of single GRU on concatenated queries (R-RE -> R-HRE) (ablation) | R-RE · Visual Genome simulated queries | R@K | Improved |
| multiple state vectors (Drill-down 5x256) instead of single 1280-d vector (R-HRE) | R-HRE · Visual Genome simulated queries | R@K | Improved |
| learned RL policy instead of fixed circular policy (ablation) | Drill-down 5x256 · Visual Genome simulated queries | R@K | Improved |
| more state vectors vs larger feature dimension at similar memory (ablation) | Drill-down / R-HRE · Visual Genome simulated queries | R@K | Improved |
| rank fusion of per-turn ranks instead of state encoding (R-RankFusion) | R-RE · Visual Genome simulated queries | R@K | Hurt |
| bidirectional query encoder instead of unidirectional (ablation) | Drill-down · Visual Genome | R@K | No clear effect |
| Drill-down 3x256 vs HRE and R-HRE baselines with real users | HRE / R-HRE · human study, 80 images | retrieval success | Improved |
Closest papers by meaning
- INQUIRE: A Natural World Text-to-Image Retrieval Benchmark (NeurIPS 2024 · similarity 0.88)
- Instance-Level Composed Image Retrieval (NeurIPS 2025 · similarity 0.86)
- Towards Natural Language-Based Document Image Retrieval: New Dataset and Benchmark (CVPR 2025 · similarity 0.86)
- SCENIR: Visual Semantic Clarity through Unsupervised Scene Graph Retrieval (ICML 2025 · similarity 0.86)
- Composite Sketch+Text Queries for Retrieving Objects with Elusive Names and Complex Interactions (AAAI 2024 · similarity 0.86)
- MagicLens: Self-Supervised Image Retrieval with Open-Ended Instructions (ICML 2024 · similarity 0.86)
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.
