Research map / Detection and segmentation / CNN features
Arbitrary-shape scene text detection
Detect curved and oriented scene text with contour, region, and localization refinement.
61 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2025) · an approach in CNN features, keyword cluster “text · event · scene”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Leveling off Several papers from 2025–26 still report on 1 tracked benchmark, but the best result has barely moved since 2024.
10 papers in 2023–24 and 2 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 3.0% to 0.6%.
Open questions the data points to
- Papers report baseline on ICDAR2015 (F-measure) anywhere from 53.3 to 85.5: comparisons against it depend on whose number you take.
- Papers report Mask TextSpotter* on IC15 (End-to-End G) anywhere from 62.4 to 73.5: comparisons against it depend on whose number you take.
What looks settled
- Total-Text (F-measure): best up to 2024 90.98, best in 2025–26 89.9 from 2 papers, at most 0.2% better, if at all.
No longer reported by papers here: ICDAR 2015 (F-measure, last in 2021); Total-Text test (F-measure, last in 2024); CTW1500 test (F-measure, last in 2024).
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2016 · Synthetic Data for Text Localisation in Natural Images (CVPR 2016 · 1,557 citations)
- 2017 · EAST: An Efficient and Accurate Scene Text Detector (CVPR 2017 · 1,852 citations)
- 2018 · PixelLink: Detecting Scene Text via Instance Segmentation (AAAI 2018 · 652 citations)
- 2019 · Shape Robust Text Detection With Progressive Scale Expansion Network (CVPR 2019 · 869 citations)
- 2020 · Real-Time Scene Text Detection with Differentiable Binarization (AAAI 2020 · 868 citations)
- 2021 · Dense Label Encoding for Boundary Discontinuity Free Rotation Detection (CVPR 2021 · 354 citations)
- 2022 · SwinTextSpotter: Scene Text Spotting via Better Synergy Between Text Detection and Text Recognition (CVPR 2022 · 166 citations)
- 2023 · DeepSolo: Let Transformer Decoder With Explicit Points Solo for Text Spotting (CVPR 2023 · 128 citations)
- 2024 · SRFormer: Text Detection Transformer with Incorporated Segmentation and Regression (AAAI 2024 · 23 citations)
- 2025 · Arbitrary Reading Order Scene Text Spotter with Local Semantics Guidance (AAAI 2025 · 8 citations)
Earliest papers here
- Synthetic Data for Text Localisation in Natural Images (CVPR 2016 · 1,557 citations)
- Multi-Oriented Text Detection With Fully Convolutional Networks (CVPR 2016 · 576 citations)
- Canny Text Detector: Fast and Robust Scene Text Localization Algorithm (CVPR 2016 · 121 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
MSRA-TD500 · F-measure (higher is better)
8 papers, 2017–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2017 | 77 | Detecting Oriented Text in Natural Images by Linking Segments (CVPR 2017 · 821 citations) |
| 2018 | 81.5 | Multi-Oriented Scene Text Detection via Corner Localization and Region Segmentation (CVPR 2018 · 395 citations) |
| 2020 | 84.9 | Real-Time Scene Text Detection with Differentiable Binarization (AAAI 2020 · 868 citations) |
| 2024 | 89.9 | CPN: Complementary Proposal Network for Unconstrained Text Detection (AAAI 2024 · 4 citations) |
| 2025 | 90.3 | Explicit Relational Reasoning Network for Scene Text Detection (AAAI 2025 · 8 citations) |
Total-Text · F-measure (higher is better)
8 papers, 2019–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 82.9 | Scene Text Detection with Supervised Pyramid Context Network (AAAI 2019 · 242 citations) |
| 2020 | 84.7 | Real-Time Scene Text Detection with Differentiable Binarization (AAAI 2020 · 868 citations) |
| 2023 | 89 | DPText-DETR: Towards Better Scene Text Detection with Dynamic Points in Transformer (AAAI 2023 · 118 citations) |
| 2024 | 90.98 | Kernel Adaptive Convolution for Scene Text Detection via Distance Map Prediction (CVPR 2024 · 20 citations) |
CTW1500 · F-measure (higher is better)
7 papers, 2020–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 83.4 | Real-Time Scene Text Detection with Differentiable Binarization (AAAI 2020 · 868 citations) |
| 2023 | 88.8 | DPText-DETR: Towards Better Scene Text Detection with Dynamic Points in Transformer (AAAI 2023 · 118 citations) |
| 2025 | 89.4 | Explicit Relational Reasoning Network for Scene Text Detection (AAAI 2025 · 8 citations) |
ICDAR 2015 · F-measure (higher is better)
7 papers, 2017–2021 · not reported since 2021
| Year | Best so far | Set by |
|---|---|---|
| 2017 | 63 | Multi-Scale FCN With Cascaded Instance Aware Segmentation for Arbitrary Oriented Word Spotting in the Wild (CVPR 2017 · 88 citations) |
| 2018 | 87.99 | FOTS: Fast Oriented Text Spotting With a Unified Network (CVPR 2018 · 581 citations) |
| 2021 | 88.2 | PGNet: Real-time Arbitrarily-Shaped Text Spotting with Point Gathering Network (AAAI 2021 · 85 citations) |
Total-Text test · F-measure (higher is better)
6 papers, 2020–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 85.4 | ContourNet: Taking a Further Step Toward Accurate Arbitrary-Shaped Scene Text Detection (CVPR 2020 · 253 citations) |
| 2021 | 85.8 | Fourier Contour Embedding for Arbitrary-Shaped Text Detection (CVPR 2021 · 303 citations) |
| 2022 | 88.1 | Few Could Be Better Than All: Feature Sampling and Grouping for Scene Text Detection (CVPR 2022 · 109 citations) |
CTW1500 test · F-measure (higher is better)
5 papers, 2020–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 83.9 | ContourNet: Taking a Further Step Toward Accurate Arbitrary-Shaped Scene Text Detection (CVPR 2020 · 253 citations) |
| 2021 | 85.5 | Fourier Contour Embedding for Arbitrary-Shaped Text Detection (CVPR 2021 · 303 citations) |
| 2024 | 86.17 | LayoutFormer: Hierarchical Text Detection Towards Scene Text Understanding (CVPR 2024 · 15 citations) |
What has been tried
61 papers here report 609 measured changes (a component added, removed or swapped, compared with a baseline): helped 78%, no clear effect 5%, hurt 15%, mixed 1%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Use 3 segmentation and 3 regression decoder layers instead of 1 segmentation and 5 regression layers | TotalText (no pre-training) · F1 | 86.5 -> 87.1 | Helped | SRFormer: Text Detection Transformer with Incorporated Segmentation and Regression (AAAI 2024 · 23 citations) |
| combine dense and sparse assignment | Total-Text · F-measure | 87.6 -> 88.1 | Helped | LRANet: Towards Accurate and Efficient Scene Text Detection with Low-Rank Approximation Network (AAAI 2024 · 21 citations) |
| add KACM to DB | TD500 · F-measure | 84.9 -> 86.1 | Helped | Kernel Adaptive Convolution for Scene Text Detection via Distance Map Prediction (CVPR 2024 · 20 citations) |
| use line rather than word or paragraph as the detection unit | HierText val · PQ | 53.09 -> 57.48; 40.34 -> 57.48 | Helped | LayoutFormer: Hierarchical Text Detection Towards Scene Text Understanding (CVPR 2024 · 15 citations) |
| Add reference point disturbance during training | Total-Text · spotting score | 77.8 -> 81.5 (+3.7) | Helped | Arbitrary Reading Order Scene Text Spotter with Local Semantics Guidance (AAAI 2025 · 8 citations) |
| Add the ERR decoder | CTW1500 · F-measure | 84.7 -> 87.2; 2.5% | Helped | Explicit Relational Reasoning Network for Scene Text Detection (AAAI 2025 · 8 citations) |
| Add Balanced RPN to Deformable MSN | IC17-MLT · F-measure | 72.7 -> 79.5 | Helped | CPN: Complementary Proposal Network for Unconstrained Text Detection (AAAI 2024 · 4 citations) |
| replace box-edge sampling with the Box-to-Poly transformation using a Bézier intermediary | InverseText, 50K iterations · F-score | 53.61 -> 87.41 | Helped | Box2Poly: Memory-Efficient Polygon Prediction of Arbitrarily Shaped and Rotated Text (AAAI 2024 · 2 citations) |
| add line-level granularity to word-level training | ICDAR2015 · F-score | 81.02 -> 89.15; 8.13-point | Helped | Towards Unified Multi-granularity Text Detection with Interactive Attention (ICML 2024 · 1 citation) |
| Insert prompts in encoder and shared feature rather than shared feature only | SCUT-EnsText · PSNR | 37.14 -> 36.93 | Hurt | UPOCR: Towards Unified Pixel-Level OCR Interface (ICML 2024 · 1 citation) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| baseline | ICDAR2015 | F-measure | 53.3 – 85.5 | 2 papers, e.g. CVPR 2018 and AAAI 2019 |
| Mask TextSpotter* | IC15 | End-to-End G | 62.4 – 73.5 | 2 papers, e.g. AAAI 2020 and AAAI 2021 |
What reviewers will expect
Datasets and baselines named in the 10 papers of this approach from 2024–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- Total-Text · 8 of 10 papers
- CTW1500 · 8 of 10 papers
- MSRA-TD500 · 6 of 10 papers
- SynthText 150K · 4 of 10 papers
- SynthText · 3 of 10 papers
- InverseText · 2 of 10 papers
- ICDAR19-ArT · 2 of 10 papers
- ICDAR2015 · 2 of 10 papers
Baselines
- DPText-DETR · 6 of 10 papers
- DB · 6 of 10 papers
- TextBPN++ · 5 of 10 papers
- ABCNet v2 · 4 of 10 papers
- FSG · 4 of 10 papers
- SwinTextSpotter · 3 of 10 papers
- I3CL · 3 of 10 papers
- PCR · 3 of 10 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 12 papers; from OpenAlex and the papers' affiliations).
- Wuhan University · 2 papers
- Tomorrow Advancing Life · 2 papers
- Baidu · 2 papers
- Fudan University · 2 papers
- University of Chinese Academy of Sciences · 2 papers
Use it
Download BibTeX of 30 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Spiking event-camera detection (40 papers)
Elsewhere in the topic
- Context modeling for semantic segmentation (241 papers)
- Region-based and query-based detectors (183 papers)
- Local features and context in CNNs (170 papers)
- Context for hard detection conditions (133 papers)
- Domain-specific segmentation and detection (118 papers)
- Convolution-transformer backbones (101 papers)
- Box regression and loss design (94 papers)
- Efficient DETR query design (85 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).