Research map / Vision-language and multimodal / Human and embodied modeling
SEED: Semantics Enhanced Encoder-Decoder Framework for Scene Text Recognition
Zhi Qiao, Yu Zhou, Dongbao Yang, Yucan Zhou, Weiping Wang
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 word embedding supervision alone (ablation) | ASTER · IC13, SVTP, and IC15 | accuracy: IC13 90.9 -> 90.8; SVTP 79.1 -> 79.2; IC15 78.4 -> 77.0 | No clear effect |
| Initialize the decoder with predicted holistic features without word embedding supervision (ablation) | ASTER · IC13, SVTP, and IC15 | accuracy: IC13 90.9 -> 91.1; SVTP 79.1 -> 78.1; IC15 78.4 -> 76.1 | Mixed |
| Combine word embedding supervision with predicted semantic decoder initialization (ablation) | ASTER · IC13, SVTP, and IC15 | accuracy: +1.9%; +2.3%; +1.6% | Improved |
| Add word embedding supervision to ASTER with semantic initialization (ablation) | ASTER + semantic initialization · IC13, SVTP, and IC15 | accuracy: +1.7%; +3.3%; +3.9% | Improved |
| Use FastText word embedding to initialize the decoder instead of predicted semantics (ablation) | SE-ASTER · Scene text recognition | accuracy | No clear effect |
| Predict semantic information from the final BiLSTM hidden state rather than the BiLSTM outputs (ablation) | SE-ASTER · Scene text recognition | accuracy | Hurt |
| Add semantic prediction and decoder initialization | SAR · IC13 | accuracy: 91.0 -> 90.9 | Hurt |
| Add semantic prediction and decoder initialization | SAR · IC15, SVT, and SVTP | accuracy: +4.2%; +1.3%; +2.3% | Improved |
| Use SE-ASTER rather than ASTER | ASTER · IC15 and SVTP | accuracy: IC15 76.1% -> 80.0%; SVTP 78.5% -> 81.4% | Improved |
| Use SE-ASTER rather than ASTER | ASTER · IIIT5K and CUTE | accuracy: IIIT5K +0.4%; CUTE +4.1% | Improved |
| Use SE-ASTER rather than ASTER | ASTER · IC13 and SVT | accuracy: IC13 92.8%; SVT 89.6% | Mixed |
| Use semantic prediction with word-embedding supervision and decoder guidance | ASTER · IC13-sr and IC15-sr | accuracy: IC13-sr 71.4 -> 77.4; IC15-sr 65.6 -> 70.0 | Improved |
Closest papers by meaning
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.87)
- Instruction-Guided Scene Text Recognition (TPAMI 2025 · similarity 0.87)
- Visual Semantics Allow for Textual Reasoning Better in Scene Text Recognition (AAAI 2022 · similarity 0.87)
- CLIPTER: Looking at the Bigger Picture in Scene Text Recognition (ICCV 2023 · 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)
- 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.
