TEACH: Text Encoding as Curriculum Hints for Scene Text Recognition
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866911088037068800 |
|---|---|
| author | Yang, Xiahan Zheng, Hui |
| author_facet | Yang, Xiahan Zheng, Hui |
| contents | Scene Text Recognition (STR) remains a challenging task due to complex visual appearances and limited semantic priors. We propose TEACH, a novel training paradigm that injects ground-truth text into the model as auxiliary input and progressively reduces its influence during training. By encoding target labels into the embedding space and applying loss-aware masking, TEACH simulates a curriculum learning process that guides the model from label-dependent learning to fully visual recognition. Unlike language model-based approaches, TEACH requires no external pretraining and introduces no inference overhead. It is model-agnostic and can be seamlessly integrated into existing encoder-decoder frameworks. Extensive experiments across multiple public benchmarks show that models trained with TEACH achieve consistently improved accuracy, especially under challenging conditions, validating its robustness and general applicability. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2508_01153 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | TEACH: Text Encoding as Curriculum Hints for Scene Text Recognition Yang, Xiahan Zheng, Hui Computer Vision and Pattern Recognition Scene Text Recognition (STR) remains a challenging task due to complex visual appearances and limited semantic priors. We propose TEACH, a novel training paradigm that injects ground-truth text into the model as auxiliary input and progressively reduces its influence during training. By encoding target labels into the embedding space and applying loss-aware masking, TEACH simulates a curriculum learning process that guides the model from label-dependent learning to fully visual recognition. Unlike language model-based approaches, TEACH requires no external pretraining and introduces no inference overhead. It is model-agnostic and can be seamlessly integrated into existing encoder-decoder frameworks. Extensive experiments across multiple public benchmarks show that models trained with TEACH achieve consistently improved accuracy, especially under challenging conditions, validating its robustness and general applicability. |
| title | TEACH: Text Encoding as Curriculum Hints for Scene Text Recognition |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2508.01153 |