Research map / Vision-language and multimodal / Human and embodied modeling
One2Seq: One-Token Wise Decoder for Efficient Scene Text Recognition
Zhibin Ma, Pengwen Dai, Wei Zhuo, Xugong Qin
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 global visual token to context token (ablation) | One2Seq OTWD · MCB avg | accuracy: 81.77 -> 82.70 | Improved |
| add global visual token (ablation) | One2Seq OTWD · CoB avg | accuracy: 96.29 -> 96.31 | No clear effect |
| add gated fusion instead of simple addition (ablation) | One2Seq DGI · U14M-B avg | accuracy: 83.98 -> 84.32 | Improved |
| add Positional-aware Hash Embedding instead of Word2Vec (ablation) | One2Seq · U14M-B avg | accuracy: 84.32 -> 85.06 | Improved |
| use one-token wise decoder instead of NRTR decoder (ablation) | One2Seq (SVTR-B) · MCB avg | accuracy: 82.92 -> 84.09 | Improved |
| use one-token wise decoder instead of NRTR decoder (ablation) | One2Seq (SVTR-B) · decoding time | ms: 23.2 -> 13.5 | Improved |
| use one-token wise decoder instead of OTE-AR (ablation) | One2Seq (SVTR-B) · decoding time | ms: 13.1 -> 13.5 (U14M-B 78.70 -> 85.06) | Mixed |
| replace SVTR-B encoder with CLIP encoder | One2Seq · Union14M-L subsets | accuracy: CUR 91.5 -> 93.2, MTO 90.8 -> 95.8, ART 79.7 -> 80.4 | Mixed |
| scale encoder from SVTR-B to SVTR-L | One2Seq · Union14M-L avg | WAIC: 85.06 -> 85.71 | Improved |
Closest papers by meaning
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.91)
- Context Perception Parallel Decoder for Scene Text Recognition (TPAMI 2025 · similarity 0.88)
- Accurate Scene Text Recognition with Efficient Model Scaling and Cloze Self-Distillation (CVPR 2025 · similarity 0.86)
- Linguistics-aware Masked Image Modeling for Self-supervised Scene Text Recognition (CVPR 2025 · similarity 0.86)
- Reading Images Like Texts: Sequential Image Understanding in Vision-Language Models (ICLR 2026 · similarity 0.85)
- Object Recognition as Next Token Prediction (CVPR 2024 · similarity 0.85)
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.
