Research map / Vision-language and multimodal / Human and embodied modeling
Multi-Granularity Prediction for Scene Text Recognition
Peng Wang, Cheng Da, Cong Yao
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 32x128 input with 4x4 patches instead of 224x224 with 16x16 (ablation) | MGP-STR ViT · 6 STR benchmarks average | average accuracy: 91.07 -> 91.58 | Improved |
| add Character A3 module instead of using first T tokens (ablation) | MGP-STR ViT · 6 STR benchmarks average | average accuracy: 91.58 -> 92.73 | Improved |
| use ViT-based vision model with A3 instead of CNN-based ABINet vision | MGP-STR_Vision vs ABINet_V · 6 STR benchmarks average | average accuracy: 89.8 -> 92.73 | Improved |
| add Cumprod fusion of character and subword heads instead of character head only (ablation) | MGP-STR_Fuse · 6 STR benchmarks average | average accuracy: 92.53 -> 93.35 | Improved |
| use Cumprod instead of Mean fusion (ablation) | MGP-STR_Fuse · 6 STR benchmarks average | average accuracy: 93.28 -> 93.35 | No clear effect |
| add BPE head to character head (ablation) | MGP-STR_C+B · 6 STR benchmarks average | average accuracy: 92.73 -> 93.24 | Improved |
| add WordPiece head to character head (ablation) | MGP-STR_C+W · 6 STR benchmarks average | average accuracy: 92.73 -> 93.25 | Improved |
| add both BPE and WordPiece heads (ablation) | MGP-STR_Fuse · 6 STR benchmarks average | average accuracy: 93.25 -> 93.35 | Improved |
| use subword head alone instead of character head (ablation) | MGP-STR_Fuse · 6 STR benchmarks average | average accuracy: 92.53 -> 85.99 | Hurt |
| add BCN language model to vision model (ablation) | MGP-STR_Vision · 6 STR benchmarks average | average accuracy: 92.40 -> 93.14 | Improved |
| use multi-granularity fusion instead of BCN explicit language model | MGP-STR_Fuse vs Vision+BCN · 6 STR benchmarks average | average accuracy: 93.14 -> 93.35 | Improved |
| add multi-granularity fusion on DeiT-Tiny (ablation) | MGP-STR · 6 STR benchmarks average (DeiT-Tiny) | average accuracy: 89.36 -> 89.91 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- RobustScanner — RobustScanner: Dynamically Enhancing Positional Clues for Robust Text Recognition (ECCV 2020) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Multi-Granularity Prediction with Learnable Fusion for Scene Text Recognition (IJCV 2026 · similarity 0.98)
- Revisiting Scene Text Recognition: A Data Perspective (ICCV 2023 · similarity 0.92)
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.89)
- An Empirical Study of Scaling Law for Scene Text Recognition (CVPR 2024 · similarity 0.88)
- Linguistics-aware Masked Image Modeling for Self-supervised Scene Text Recognition (CVPR 2025 · similarity 0.87)
- Image as a Language: Revisiting Scene Text Recognition via Balanced, Unified and Synchronized Vision-Language Reasoning Network (AAAI 2024 · 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.
