Research map / Vision-language and multimodal / Human and embodied modeling
Visual Semantics Allow for Textual Reasoning Better in Scene Text Recognition
Read the paperCheck 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 |
|---|---|---|---|
| Add GTR to the segmentation-based VR model (ablation) | Segmentation-based VR · STR benchmarks (average across six test sets) | word accuracy: +4.06% | Helped |
| Add LM to the segmentation-based VR model (ablation) | Segmentation-based VR · STR benchmarks (average across six test sets) | word accuracy: +3.45% | Helped |
| Use both LM and GTR instead of VR alone (ablation) | Segmentation-based VR · STR benchmarks (average across six test sets) | word accuracy: 90.96% | Helped |
| Add mutual learning between LM and GTR (ablation) | S-GTR without mutual learning · STR benchmarks (average across six test sets) | word accuracy: 91.92% | Helped |
| Plug GTR into CRNN | CRNN · STR test sets | average word accuracy: +3.77% | Helped |
| Plug GTR into TRBA | TRBA · STR test sets | average word accuracy: +3.20% | Helped |
| Plug GTR into SRN | SRN · STR test sets | average word accuracy: +2.78% | Helped |
| Plug GTR into Base2D | Base2D · STR test sets | average word accuracy: +1.69% | Helped |
| Plug GTR into ABINet-LV | ABINet-LV · STR test sets | average word accuracy: +1.65% | Helped |
| Use 3 first-stage GCN layers instead of 2 (ablation) | GTR · IIIT5k | word accuracy (%): 95.8 -> 96.0 | Helped |
| Use 1 first-stage GCN layer instead of 2 (ablation) | GTR · IIIT5k | word accuracy (%): 95.8 -> 94.3 | Hurt |
| Use continuous adjacency values instead of discrete values (ablation) | GTR · IIIT5k | word accuracy (%): 95.8 -> 95.9 | No clear effect |
Closest papers by meaning
- Image as a Language: Revisiting Scene Text Recognition via Balanced, Unified and Synchronized Vision-Language Reasoning Network (AAAI 2024 · similarity 0.90)
- Linguistics-aware Masked Image Modeling for Self-supervised Scene Text Recognition (CVPR 2025 · similarity 0.89)
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.88)
- Scene Graph Thinking: Reinforcing Structured Visual Reasoning for Multimodal Large Language Models (ICML 2026 · similarity 0.88)
- Aligning Forest and Trees in Images & Long Captions for Visually Grounded Understanding (NeurIPS 2026 · similarity 0.88)
- VLM-R³: Region Recognition, Reasoning, and Refinement for Enhanced Multimodal Chain-of-Thought (NeurIPS 2025 · similarity 0.88)
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.