TABLE 2.
Classification accuracies of ResNet50, InceptionV3 and InceptionResNetV2 models on images with different resolution.
| Resolution |
||||||
| 197 × 197 |
224 × 224 |
448 × 448 |
||||
| Backbone network | Accuracy (%) | FPS | Accuracy (%) | FPS | Accuracy (%) | FPS |
| Inceptionv3 | 91.70 ± 0.17 | 135.4 | 91.72 ± 0.16 | 67.7 | 92.73 ± 0.13 | 31.3 |
| ResNet50 | 89.97 ± 0.24 | 126.1 | 90.23 ± 0.15 | 63.3 | 91.48 ± 0.12 | 29.0 |
| InceptionResNetV2 | 91.09 ± 0.23 | 101.6 | 91.48 ± 0.12 | 50.6 | 92.99 ± 0.15 | 22.6 |
Frames per second (FPS) indicate the speed. Average accuracies ± standard deviations were calculated from 5-fold cross validation of the BreaKHis dataset.