ICDAR 2015
Unknown
1000 training + 500 test images captured with wearable cameras. Industry standard for scene text detection.
Benchmark Stats
Models4
Papers4
Metrics1
SOTA History
Only 4 models on this benchmark
Help build the community leaderboard — submit your model results.
f-measure
Higher is better
| Rank | Model | Source | Score | Year | Paper |
|---|---|---|---|---|---|
| 1 | TESTR TESTR (Text Spotting Transformers), CVPR 2022. Detection-only F-measure on ICDAR 2015 test set without lexicon. arxiv:2204.01918. | Community | 90 | 2022 | Source |
| 2 | FreeReal+DBNet FreeReal with DBNet backbone, ECCV 2024. Bridging synthetic and real worlds for pre-training. Achieves 90.0% F-measure on IC15. arxiv:2312.05286. | Community | 90 | 2024 | Source |
| 3 | EK-Net++ EK-Net++ improves EK-Net with Epoch Adaptive Weight algorithm. Expert Systems with Applications 2024. | Community | 86.72 | 2024 | Source |
| 4 | EK-Net EK-Net (Expand Kernel Network), arXiv Jan 2024. ResNet-18 backbone at 35.42 FPS. arxiv:2401.11704. | Community | 85.72 | 2024 | Source |