Research map / Architectures and efficiency / Neural architecture search
Searching a High Performance Feature Extractor for Text Recognition Network
Hui Zhang, Quanming Yao, James Tin-Yau Kwok, 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 |
|---|---|---|---|
| use TREFE instead of AutoSTR | AutoSTR · IAM | WER: 45.23 -> 16.41 | Improved |
| use TREFE instead of AutoSTR | AutoSTR · RIMES | WER: 20.40 -> 9.16 | Improved |
| use TREFE instead of AutoSTR (latency) | AutoSTR · Scene text | latency (ms): 3.86 -> 2.62 | Improved |
| use TREFE instead of AutoSTR | AutoSTR · IC15 (1811) | word accuracy: 81.7 -> 84.0 | Improved |
| add searched spatial model instead of ResNet (ablation) | TREFE with searched sequential model · IAM | WER: 18.67 -> 16.41 | Improved |
| add searched spatial model instead of VGG (ablation) | TREFE with searched sequential model · IAM | WER: 20.80 -> 16.41 | Improved |
| use searched spatial model instead of DARTS CNN (ablation) | TREFE with searched sequential model · IAM | WER: 23.24 -> 16.41 | Improved |
| use searched sequential model instead of BiLSTM (ablation) | TREFE with searched spatial model · IAM | WER: 18.65 -> 16.41 | Improved |
| use searched sequential model instead of vanilla Transformer (ablation) | TREFE with searched spatial model · IAM | WER: 19.84 -> 16.41 | Improved |
| use searched sequential model instead of Evolved Transformer (ablation) | TREFE with searched spatial model · IAM | WER: 20.27 -> 16.41 | Improved |
| use NAS-searched TREFE instead of random search, no latency constraint | random search · IAM | WER: 17.04 -> 16.41 | Improved |
| use TREFE instead of random search at 2/3 latency | random search · IAM | WER: 19.39 -> 19.18 | No clear effect |
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.
- AutoSTR — AutoSTR: Efficient Backbone Search for Scene Text Recognition (ECCV 2020) · 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
- TextNAS: A Neural Architecture Search Space Tailored for Text Representation (AAAI 2020 · similarity 0.89)
- AutoSTR: Efficient Backbone Search for Scene Text Recognition (ECCV 2020 · similarity 0.87)
- TextAdaIN: Paying Attention to Shortcut Learning in Text Recognizers (ECCV 2022 · similarity 0.87)
- An End-to-End TextSpotter With Explicit Alignment and Attention (CVPR 2018 · similarity 0.85)
- einspace: Searching for Neural Architectures from Fundamental Operations (NeurIPS 2024 · similarity 0.84)
- Out of Length Text Recognition with Sub-String Matching (AAAI 2025 · similarity 0.83)
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.
