End-to-End Recognition on ICDAR 2015 (test)
35Strong Error RateNeumann et al.
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Neumann et al.Inference Scale=Single Scale2019.10 | 35 | 20 | 16 | — | |
| TextProp.+DictNetInference Scale=Single Scale2019.10 | 53.3 | 49.61 | 47.18 | — | |
| Deep text spotterInference Scale=Single Scale2019.10 | 54 | 51 | 47 | — | |
| Mask TextSpotterInference Scale=Single Scale2019.10 | 79.3 | 73 | 62.4 | — | |
| CharNet R-50Inference Scale=Single Scale, Backbone=ResNet-502019.10 | 80.14 | 74.45 | 62.18 | 60.72 | |
| FOTS R-50Inference Scale=Single Scale, Backbone=ResNet-502019.10 | 81.09 | 75.9 | 60.8 | — | |
| FOTS2022.03 | 81.1 | 75.9 | 60.8 | — | |
| CharNet H-57Inference Scale=Single Scale, Backbone=Hourglass-572019.10 | 81.43 | 77.62 | 66.92 | 62.79 | |
| MANGO2022.03 | 81.8 | 78.9 | 67.3 | — | |
| He et al. MSInference Scale=Multi-Scale2019.10 | 82 | 77 | 63 | — | |
| CharNet R-50 MSInference Scale=Multi-Scale, Backbone=ResNet-502019.10 | 82.46 | 78.86 | 67.64 | 62.71 | |
| TextDragon2022.03 | 82.5 | 78.3 | 65.2 | — | |
| PAN++2022.03 | 82.7 | 78.2 | 69.2 | — | |
| ABCNet v22022.03 | 82.7 | 78.5 | 73 | — | |
| Mask TextSpotter2022.03 | 83 | 77.7 | 73.5 | — | |
| CharNet H-88Inference Scale=Single Scale, Backbone=Hourglass-882019.10 | 83.1 | 79.15 | 69.14 | 65.73 | |
| CharNet2022.03 | 83.1 | 79.2 | 69.1 | — | |
| Mask TextSpotter v32022.03 | 83.3 | 78.1 | 74.2 | — | |
| FOTS R-50 MSInference Scale=Multi-Scale, Backbone=ResNet-502019.10 | 83.55 | 79.11 | 65.33 | — | |
| SwinTextSpotter2022.03 | 83.9 | 77.3 | 70.5 | — | |
| CharNet H-57 MSInference Scale=Multi-Scale, Backbone=Hourglass-572019.10 | 84.07 | 80.1 | 69.21 | 65.26 | |
| CharNet H-88 MSInference Scale=Multi-Scale, Backbone=Hourglass-882019.10 | 85.05 | 81.25 | 71.08 | 67.24 |