- Task 2 - Script identification - Method: CNN-Based Classifier
- Method info
- Samples list
- Per sample details
method: CNN-Based Classifier2019-06-03
Authors: Hongzhen Wang, Mengkai Ma, Shan Huang, Qiufeng Wang, Kaizhu Huang, Affiliation: SOT Research
Description: We used ResNet as the backbone network, and applied some simple data augmentations.
Confusion Matrix
| Detection | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Arabic | Latin | Chinese | Japanese | Korean | Bangla | Hindi | Symbols | None | ||
| GT | Arabic | 4846 | 219 | 1 | 19 | 21 | 8 | 3 | 18 | 7 |
| Latin | 224 | 59087 | 130 | 385 | 424 | 116 | 71 | 186 | 14 | |
| Chinese | 8 | 164 | 3664 | 825 | 67 | 9 | 1 | 12 | 0 | |
| Japanese | 41 | 1342 | 1052 | 5357 | 298 | 12 | 37 | 18 | 0 | |
| Korean | 33 | 953 | 143 | 309 | 11515 | 15 | 14 | 10 | 0 | |
| Bangla | 6 | 133 | 8 | 8 | 12 | 2251 | 124 | 0 | 3 | |
| Hindi | 8 | 40 | 0 | 3 | 3 | 25 | 4134 | 5 | 6 | |
| Symbols | 86 | 782 | 4 | 39 | 26 | 9 | 9 | 3059 | 1 | |
| None | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | |