Research map / Vision-language and multimodal / Human and embodied modeling
SVTRv2: CTC Beats Encoder-Decoder Models in Scene Text Recognition
Yongkun Du, Zhineng Chen, Hongtao Xie, Caiyan Jia, Yu-Gang Jiang
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 multi-size resizing instead of fixed 32x128 resizing (ablation) | SVTRv2 · U14M irregular text (R1 aspect-ratio group) | word accuracy: 72.1 -> 87.4 | Improved |
| add multi-size resizing instead of fixed 32x128 resizing (ablation) | SVTRv2 · U14M irregular text (R2 aspect-ratio group) | word accuracy: 83.1 -> 88.3 | Improved |
| use multi-size resizing instead of larger fixed 64x256 (ablation) | SVTRv2 · U14M avg | word accuracy: 79.03 -> 83.86 | Improved |
| add horizontal feature rearranging (ablation) | SVTRv2 · U14M Multi-Oriented | word accuracy: 83.73 -> 85.76 | Improved |
| add vertical feature rearranging (ablation) | SVTRv2 · U14M Multi-Oriented | word accuracy: 83.73 -> 84.44 | Improved |
| add full feature rearranging module (H and V) (ablation) | SVTRv2 · U14M Multi-Oriented | word accuracy: 83.73 -> 86.19 | Improved |
| replace FRM with one Transformer block (ablation) | SVTRv2 · U14M avg | word accuracy: 82.94 -> 82.49 | Hurt |
| add semantic guidance module (ablation) | SVTRv2 · OST avg | word accuracy: 74.92 -> 80.03 | Improved |
| add semantic guidance module (ablation) | SVTRv2 · U14M avg | word accuracy: 83.86 -> 86.14 | Improved |
| add semantic guidance module (ablation) | SVTRv2 · Com avg | word accuracy: 96.16 -> 96.57 | Improved |
| use SGM instead of PARSeq language model during training (ablation) | SVTRv2 · OST avg | word accuracy: 76.55 -> 80.03 | Improved |
| remove rectification module TPS from SVTR (ablation) | SVTR · LTB long text | word accuracy: 0.0 -> 45.1 | Improved |
Closest papers by meaning
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.89)
- Revisiting Scene Text Recognition: A Data Perspective (ICCV 2023 · similarity 0.88)
- What Is Wrong With Scene Text Recognition Model Comparisons? Dataset and Model Analysis (ICCV 2019 · similarity 0.88)
- Multi-Granularity Prediction with Learnable Fusion for Scene Text Recognition (IJCV 2026 · similarity 0.88)
- Multi-Granularity Prediction for Scene Text Recognition (ECCV 2022 · similarity 0.87)
- Linguistics-aware Masked Image Modeling for Self-supervised Scene Text Recognition (CVPR 2025 · 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.
