Research map / Detection and segmentation / CNN features
Towards End-To-End Text Spotting With Convolutional Recurrent Neural Networks
Hui Li, Peng Wang, Chunhua Shen
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 attention to the word decoder (ablation) | Ours Atten+Fixed · ICDAR2015 End-to-End (Generic lexicon) | F-measure (%): 79.18 -> 83.98 | Improved |
| add attention to the word decoder (ablation) | Ours Atten+Fixed · SVT Word-Spotting (Strong lexicon) | F-measure (%): 79.49 -> 81.80 | Improved |
| use varying-size RoI pooling instead of fixed 4x20 (ablation) | Ours Atten+Vary · ICDAR2011 Word-Spotting (Generic lexicon) | F-measure (%): 83.81 -> 87.70 | Improved |
| use varying-size RoI pooling instead of fixed 4x20 (ablation) | Ours Atten+Vary · SVT Word-Spotting (Strong lexicon) | F-measure (%): 81.80 -> 84.91 | Improved |
| use varying-size RoI pooling instead of fixed 4x20 (ablation) | Ours Atten+Vary · ICDAR2015 Word-Spotting (Strong lexicon) | F-measure (%): 93.33 -> 94.16 | Improved |
| train detection and recognition jointly instead of two separate stages (ablation) | Ours Atten+Vary · ICDAR2015 End-to-End (Strong lexicon) | F-measure (%): 88.20 -> 91.08 | Improved |
| train detection and recognition jointly instead of two separate stages (ablation) | Ours Atten+Vary · ICDAR2011 Word-Spotting (Generic lexicon) | F-measure (%): 82.86 -> 87.70 | Improved |
| add recognition loss to detector training (joint multi-task) (ablation) | Ours Atten+Vary · ICDAR2015 text detection | F-measure (%): 83.4 -> 85.6 | Improved |
| add recognition loss to detector training (joint multi-task) (ablation) | Ours Atten+Vary · ICDAR2011 text detection | F-measure (%): 83.5 -> 85.1 | Improved |
| share convolutional features between detection and recognition (ablation) | Ours Atten+Vary · 600x800 image, M40 GPU | recognition time: 0.45s -> 0.15s | Improved |
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.
- TextBoxes — TextBoxes: A Fast Text Detector with a Single Deep Neural Network (AAAI 2017) · better on 5 of 6 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
- Towards End-to-End Text Spotting in Natural Scenes (TPAMI 2021 · similarity 0.93)
- An End-to-End TextSpotter With Explicit Alignment and Attention (CVPR 2018 · similarity 0.92)
- Mask TextSpotter: An End-to-End Trainable Neural Network for Spotting Text with Arbitrary Shapes (ECCV 2018 · similarity 0.92)
- SEE: Towards Semi-Supervised End-to-End Scene Text Recognition (AAAI 2018 · similarity 0.90)
- Detecting Text in Natural Image with Connectionist Text Proposal Network (ECCV 2016 · similarity 0.90)
- Bridging the Gap Between End-to-End and Two-Step Text Spotting (CVPR 2024 · similarity 0.88)
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.
