Research map / Detection and segmentation / CNN features
GTC: Guided Training of CTC towards Efficient and Accurate Scene Text Recognition
Wenyang Hu, Xiaocong Cai, Jun Hou, Shuai Yi, Zhiping Lin
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 |
|---|---|---|---|
| Train CTC using attention guidance rather than using CTC to guide attention (ablation) | GTC · All public benchmarks | overall accuracy (%): 87.87% -> 90.06% | Improved |
| Add attentional guidance to the CTC model | CTC Baseline · IIIT5K | accuracy (%): 95.4 -> 96.0 | Improved |
| Add attentional guidance to the CTC model | CTC Baseline · IC03 | accuracy (%): 93.6 -> 95.8 | Improved |
| Add attentional guidance to the CTC model | CTC Baseline · IC15 | accuracy (%): 76.4 -> 79.5 | Improved |
| Add a GCN to CTC without guided training (ablation) | CTC Baseline · CUTE80 | accuracy (%): 85.7 -> 88.2 | Improved |
| Add a GCN to GTC (1D) (ablation) | GTC (1D) · SVT | accuracy (%): 91.8 -> 93.2 | Improved |
| Add a GCN to GTC (2D) (ablation) | GTC (2D) · IC13 | accuracy (%): 92.6 -> 94.4 | Improved |
| Use 2D attentional guidance instead of 1D guidance (ablation) | GTC without GCN · IIIT5K | accuracy (%): 96.0 -> 95.0 | Hurt |
| Use 2D attentional guidance instead of 1D guidance (ablation) | GTC without GCN · IC13 | accuracy (%): 93.2 -> 92.6 | No clear effect |
| Use 2D attentional guidance instead of 1D guidance (ablation) | GTC + GCN · CUTE80 | accuracy (%): 92.0 -> 92.2 | Improved |
| Use CTC loss to guide the attention decoder instead of attention guidance for CTC (ablation) | Guided training · All public benchmarks | overall accuracy (%): 90.06% -> 87.87% | Hurt |
| Use the GCN+CTC decoder instead of the attention decoder at inference | GTC · Inference test | inference time (ms/image): 54 -> 3.7 | Improved |
Closest papers by meaning
- SVTRv2: CTC Beats Encoder-Decoder Models in Scene Text Recognition (ICCV 2025 · similarity 0.87)
- Connectionist Temporal Classification with Maximum Entropy Regularization (NeurIPS 2018 · similarity 0.86)
- Self-Distillation Regularized Connectionist Temporal Classification Loss for Text Recognition: A Simple Yet Effective Approach (AAAI 2024 · similarity 0.85)
- Detecting Text in Natural Image with Connectionist Text Proposal Network (ECCV 2016 · similarity 0.84)
- Reading Scene Text in Deep Convolutional Sequences (AAAI 2016 · similarity 0.84)
- Primitive Representation Learning for Scene Text Recognition (CVPR 2021 · 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.
