Research map / Detection and segmentation / CNN features
Towards End-to-End Text Spotting in Natural Scenes
Peng Wang, Hui Li, 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 joint training of detection and recognition (ablation) | Ours (sep) · ICDAR2015 End-to-End Strong | F-measure (%): 79.86 -> 82.21 | Improved |
| add joint training (ablation) | Ours (sep) · Total-Text End-to-End | F-measure (%): 57.82 -> 58.72 | Improved |
| use varying-size RoI pooling instead of fixed 4x20 (ablation) | Ours · ICDAR2015 End-to-End Strong | F-measure (%): 82.12 -> 82.21 | Improved |
| use varying-size RoI pooling instead of fixed 7x7 (ablation) | Ours · Total-Text End-to-End | F-measure (%): 50.82 -> 58.72 | Improved |
| use varying-size instead of fixed 4x30 RoI pooling (ablation) | Ours · ICDAR2015 End-to-End Strong | F-measure (%): 81.72 -> 82.21 | Improved |
| use LSTM RoI encoding instead of average pooling (ablation) | Ours · Total-Text End-to-End | F-measure (%): 55.95 -> 58.72 | Improved |
| add FC layer after average pooling (ablation) | Ours (AP) · Total-Text End-to-End | F-measure (%): 55.95 -> 57.06 | Improved |
| use 2D attention instead of 1D attention (ablation) | Ours (1D) · ICDAR2015 End-to-End Strong | F-measure (%): 77.40 -> 78.75 | Improved |
| add box refinement (ablation) | Ours (2D) · Total-Text End-to-End | F-measure (%): 50.39 -> 58.20 | Improved |
| add box refinement (ablation) | Ours (2D) · ICDAR2013 End-to-End Strong | F-measure (%): 92.61 -> 92.61 | No clear effect |
| fit 6-point polygons instead of quadrangles (ablation) | Ours_Quad · Total-Text End-to-End | F-measure (%): 58.72 -> 59.11 | Improved |
| separate LSTM encoders for detection and recognition instead of sharing (ablation) | Ours (Shr) · ICDAR2015 End-to-End Strong | F-measure (%): 82.43 -> 82.21 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 3 on most shared results.
- FOTS — FOTS: Fast Oriented Text Spotting With a Unified Network (CVPR 2018) · better on 5 of 6 shared results
- TextBoxes++ — TextBoxes: A Fast Text Detector with a Single Deep Neural Network (AAAI 2017) · better on 4 of 6 shared results
- TextDragon — TextDragon: An End-to-End Framework for Arbitrary Shaped Text Spotting (ICCV 2019) · better on 0 of 1 shared result
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 With Convolutional Recurrent Neural Networks (ICCV 2017 · similarity 0.93)
- An End-to-End TextSpotter With Explicit Alignment and Attention (CVPR 2018 · similarity 0.91)
- Towards Unconstrained End-to-End Text Spotting (ICCV 2019 · similarity 0.90)
- Mask TextSpotter: An End-to-End Trainable Neural Network for Spotting Text with Arbitrary Shapes (ECCV 2018 · similarity 0.90)
- Character Region Attention For Text Spotting (ECCV 2020 · 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.
