Research map / Vision-language and multimodal / Human and embodied modeling
Read Like Humans: Autonomous, Bidirectional and Iterative Language Modeling for Scene Text Recognition
Check 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 |
|---|---|---|---|
| replace unidirectional or ensemble SRN language modeling with BCN (ablation) | SV and LV vision models · IC13, SVT, IIIT, IC15, SVTP, CUTE80 | accuracy: 90.6 -> 91.0; 91.9 -> 92.3 | Helped |
| block gradient flow between vision and language models (ablation) | ABINet · six STR benchmarks, Avg | accuracy: 91.4 -> 92.3 | Helped |
| pre-train the vision model on MJ and ST (ablation) | ABINet · six STR benchmarks, Avg | accuracy: 91.7 -> 92.3 | Helped |
| pre-train the language model on MJ and ST text (ablation) | ABINet · six STR benchmarks, Avg | accuracy: 91.7 -> 91.6 | No clear effect |
| pre-train the language model on WikiText-103 (ablation) | ABINet · six STR benchmarks, Avg | accuracy: 92.3 -> 92.5 | Helped |
| increase BCN iterations from 1 to 3 (ablation) | SV + BCN · six STR benchmarks, Avg | accuracy: 91.0 -> 91.4 | Helped |
| increase BCN iterations from 1 to 3 (ablation) | LV + BCN · six STR benchmarks, Avg | accuracy: 92.3 -> 92.6 | Helped |
| increase BCN iterations (ablation) | ABINet · inference | inference time: 19.5 -> 31.6; 22 -> 33.9 | Hurt |
| use position attention and a deeper Transformer vision model (ablation) | vision model · six STR benchmarks, Avg | accuracy: 88.8 -> 89.8 | Helped |
| add unlabeled Uber-Text self-training | ABINet-LV · IC13, SVT, IIIT, IC15, SVTP, CUTE80 | accuracy: 97.4 -> 97.3; 93.5 -> 94.9; 96.2 -> 96.8; 86.0 -> 87.4; 89.3 -> 90.1; 89.2 -> 93.4 | Helped |
| use ensemble self-training instead of standard self-training | ABINet-LV with Uber-Text · IC13, SVT, IIIT, IC15, SVTP, CUTE80 | accuracy: 97.3 -> 97.7; 94.9 -> 95.5; 96.8 -> 97.2; 87.4 -> 86.9; 90.1 -> 89.9; 93.4 -> 94.1 | Mixed |
| replace the reproduced SRN with ABINet-LV | LV vision model · IC13, SVT, IIIT, IC15, SVTP, CUTE80 | accuracy: 96.8 -> 97.4; 92.3 -> 93.5; 96.3 -> 96.2; 84.2 -> 86.0; 87.9 -> 89.3; 88.2 -> 89.2 | Mixed |
Closest papers by meaning
- Image as a Language: Revisiting Scene Text Recognition via Balanced, Unified and Synchronized Vision-Language Reasoning Network (AAAI 2024 · similarity 0.89)
- Accurate Scene Text Recognition with Efficient Model Scaling and Cloze Self-Distillation (CVPR 2025 · similarity 0.87)
- Linguistics-aware Masked Image Modeling for Self-supervised Scene Text Recognition (CVPR 2025 · similarity 0.86)
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.85)
- DINOv2 Meets Text: A Unified Framework for Image- and Pixel-Level Vision-Language Alignment (CVPR 2025 · similarity 0.85)
- Vision‑Language‑Vision Auto‑Encoder: Scalable Knowledge Distillation from Diffusion Models (NeurIPS 2025 · similarity 0.84)
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.