Research map / Architectures and efficiency / Neural architecture search
AutoSTR: Efficient Backbone Search for Scene Text Recognition
Hui Zhang, Quanming Yao, Mingkun Yang, Yongchao Xu, Xiang Bai
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 searched backbone (AutoSTR) instead of ASTER's fixed backbone | ASTER pipeline · IIIT5K | word accuracy: 93.3 -> 94.7 | Improved |
| add searched backbone (AutoSTR) instead of ASTER's fixed backbone | ASTER pipeline · IC15 | word accuracy: 78.5 -> 81.8 | Improved |
| add searched backbone (AutoSTR) instead of ASTER's fixed backbone | ASTER pipeline · IC13 | word accuracy: 91.5 -> 94.2 | Improved |
| add per-layer operation search instead of all MBConv(k:5,e:6) (ablation) | AutoSTR · IC13 | word accuracy: 92.3 -> 94.2 | Improved |
| add per-layer operation search instead of all MBConv(k:5,e:6) (ablation) | AutoSTR · IIIT5K | word accuracy: 94.5 -> 94.7 | Improved |
| add AutoSTR search instead of reusing DARTS cell (ablation) | backbone · IIIT5K | word accuracy: 90.6 -> 94.7 | Improved |
| add AutoSTR search instead of reusing AutoDeepLab cell (ablation) | backbone · SVT | word accuracy: 87.2 -> 90.9 | Improved |
| use ABABB downsampling path instead of AABBB (ablation) | AutoSTR operation search · IIIT5K | word accuracy: 93.9 -> 94.7 | Improved |
| add operation search (step 2) after path search with ABABB (ablation) | AutoSTR · IIIT5K | word accuracy: 93.1 -> 94.7 | Improved |
| raise FLOPS regularizer beta from 0.0 to 0.9 (ablation) | AutoSTR · IIIT5K | word accuracy and FLOPS: 94.6 -> 93.5 accuracy, 319 -> 149 FLOPS (M) | Mixed |
| raise FLOPS regularizer beta from 0.0 to 0.6 (ablation) | AutoSTR · IIIT5K | word accuracy: 94.6 -> 94.7 accuracy, 319 -> 256 FLOPS (M) | No clear effect |
| use AutoSTR search instead of random architectures from the same space (ablation) | AutoSTR · IIIT5K | word accuracy: 0.5%-1.4% | Improved |
Later papers that compared with this method
1 later accepted paper reports AutoSTR as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Searching a High Performance Feature Extractor for Text Recognition Network (TPAMI 2022 · better on 9 of 9 shared results)
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
- Revisiting Scene Text Recognition: A Data Perspective (ICCV 2023 · similarity 0.88)
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.87)
- Searching a High Performance Feature Extractor for Text Recognition Network (TPAMI 2022 · similarity 0.87)
- What Is Wrong With Scene Text Recognition Model Comparisons? Dataset and Model Analysis (ICCV 2019 · similarity 0.87)
- MSTAR: Box-free Multi-query Scene Text Retrieval with Attention Recycling (NeurIPS 2025 · similarity 0.85)
- Accurate Scene Text Recognition with Efficient Model Scaling and Cloze Self-Distillation (CVPR 2025 · similarity 0.85)
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.
