Research map / Vision-language and multimodal / Human and embodied modeling
Single Shot Scene Text Retrieval
Lluis Gomez, Andres Mafla, Marcal Rusinol, Dimosthenis Karatzas
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 Annoy approximate NN instead of exact ball-tree search (ablation) | Proposed 608x608 · IIIT-STR | mAP / search time (s): 0.6983 -> 0.6883 mAP; 0.4321 -> 0.0027 secs | No clear effect |
| use HNSW approximate NN instead of exact ball-tree search (ablation) | Proposed 608x608 · IIIT-STR | mAP / search time (s): 0.6983 -> 0.6922 mAP; 0.4321 -> 0.0018 secs | No clear effect |
| use HNSW approximate NN instead of exact ball-tree search (ablation) | Proposed 608x608 · Sports-10K | mAP / search time (s): 0.7375 -> 0.7247 mAP; 0.6826 -> 0.0223 secs | No clear effect |
| use Falconn LSH instead of exact ball-tree search (ablation) | Proposed 608x608 · Sports-10K | mAP / search time (s): 0.7375 -> 0.7201 mAP; 0.6826 -> 0.0178 secs | No clear effect |
| increase input size from 576x576 to 608x608 (ablation) | Proposed · IIIT-STR | mAP: 68.13 -> 69.83 (fps 53.0 -> 43.5) | Improved |
| combine outputs at three resolutions (544, 576, 608) (ablation) | Proposed · IIIT-STR | mAP: 69.83 -> 71.37 (fps 43.5 -> 16.1) | Improved |
| combine outputs at three resolutions (544, 576, 608) (ablation) | Proposed · SVT | mAP: 83.74 -> 85.18 | Improved |
| query words unseen in training instead of seen ones | Proposed · IIIT-STR | mAP: 69.14 -> 74.92 | No clear effect |
| use PHOC space for look-up in end-to-end recognition (He et al.) instead of direct matching | He et al. · IIIT-STR | mAP: 50.16 -> 46.34 | Hurt |
| add a predefined dictionary to filter He et al. results | He et al. · Sports-10K | mAP: 50.74 -> 74.27 | Improved |
| count unannotated occluded instances of castrol (manual inspection) | Proposed · Sports-10K (castrol query) | P@10: 50 -> 95.0 | Improved |
Closest papers by meaning
- Unambiguous Text Localization and Retrieval for Cluttered Scenes (CVPR 2017 · similarity 0.87)
- MSTAR: Box-free Multi-query Scene Text Retrieval with Attention Recycling (NeurIPS 2025 · similarity 0.86)
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.86)
- Unambiguous Text Localization, Retrieval, and Recognition for Cluttered Scenes (TPAMI 2020 · similarity 0.86)
- Scene Text Retrieval via Joint Text Detection and Similarity Learning (CVPR 2021 · similarity 0.86)
- SIFT Meets CNN: A Decade Survey of Instance Retrieval (TPAMI 2017 · 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.
