Table 1. Test accuracy of patch-based classification averaged over two runs (for two subsets described in Experimental setup).
| Method | CA | CG | CL | CN | CP | CT | MF | SB | SC | BG | Total | Training time (s) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| AlexNet | 78.6 ± 4.3 | 80.0 ± 1.4 | 55.8 ± 1.4 | 63.4 ± 14.7 | 75.0 ± 7.9 | 35.0 ± 6.4 | 71.6 ± 16.6 | 67.9 ± 5.0 | 72.1 ± 2.1 | 90.0 ± 0.1 | 71.6 ± 2.4 | 250600 |
| DenseNet169 | 67.9 ± 17.9 | 72.1 ± 0.7 | 53.6 ± 0.7 | 56.3 ± 13.3 | 60.0 ± 12.7 | 81.4 ± 0.7 | 68.7 ± 5.6 | 85.0 ± 5.0 | 68.8 ± 8.6 | 81.2 ± 2.0 | 72.9 ± 0.6 | 271600 |
| InceptionV3 | 67.3 ± 1.5 | 55.0 ± 2.2 | 59.3 ± 6.4 | 67.0 ± 18.2 | 64.3 ± 1.4 | 81.4 ± 2.9 | 61.1 ± 11.1 | 55.0 ± 0.7 | 89.3 ± 1.3 | 84.7 ± 3.1 | 69.9 ± 1.9 | 309400 |
| ResNet18 | 91.4 ± 3.6 | 67.1 ± 5.7 | 67.1 ± 5.7 | 61.9 ± 7.5 | 65.0 ± 0.7 | 69.8 ± 2.8 | 54.8 ± 12.8 | 93.6 ± 3.2 | 93.5 ± 1.2 | 93.1 ± 0.5 | 75.9 ± 2.6 | 263900 |
| ResNet50 | 86.4 ± 3.5 | 57.9 ± 2.1 | 89.3 ± 3.5 | 61.8 ± 11.5 | 60.0 ± 7.1 | 64.5 ± 6.0 | 66.4 ± 19.3 | 79.3 ± 0.7 | 64.3 ± 12.4 | 90.4 ± 2.7 | 73.9 ± 2.6 | 276500 |
| AlexNet BoW RF | 87.3 ± 9.7 | 39.0 ± 0.3 | 88.3 ± 7.0 | 64.0 ± 17.0 | 79.0 ± 11.0 | 79.3 ± 10.7 | 55.1 ± 3.3 | 93.3 ± 2.0 | 81.3 ± 6.0 | 92.6 ± 1.0 | 76.7 ± 1.0 | 156 |
| InceptionV3 BoW RF | 44.3 ± 7.7 | 51.0 ± 13.7 | 60.3 ± 0.3 | 34.2 ± 17.8 | 28.0 ± 0.7 | 40.7 ± 11.3 | 13.3 ± 1.4 | 39.7 ± 11.7 | 36.3 ± 8.3 | 83.5 ± 2.0 | 44.6 ± 0.3 | 155 |
| ResNet18 BoW RF | 54.3 ± 2.3 | 39.7 ± 8.3 | 80.7 ± 12.7 | 46.2 ± 20.1 | 72.0 ± 3.3 | 61.3 ± 11.3 | 42.1 ± 2.9 | 58.7 ± 8.7 | 68.0 ± 4.0 | 91.6 ± 2.0 | 62.7 ± 2.4 | 158 |
| AlexNet BoW SVM | 92.3 ± 3.0 | 44.3 ± 18.3 | 88.7 ± 9.3 | 62.2 ± 22.2 | 83.0 ± 5.7 | 71.0 ± 15.0 | 76.1 ± 11.8 | 85.0 ± 4.3 | 76.3 ± 7.0 | 89.5 ± 3.6 | 77.6 ± 1.2 | 5124 |
| InceptionV3 BoW SVM | 44.3 ± 10.3 | 32.3 ± 5.7 | 57.7 ± 2.3 | 33.3 ± 17.6 | 28.3 ± 1.0 | 39.7 ± 6.3 | 13.0 ± 1.7 | 34.7 ± 8.0 | 32.0 ± 2.0 | 78.8 ± 2.1 | 40.8 ± 1.3 | 5153 |
| ResNet18 BoW SVM | 59.0 ± 3.7 | 32.3 ± 5.7 | 84.0 ± 11.3 | 54.7 ± 22.0 | 66.0 ± 0.0 | 62.0 ± 4.0 | 39.7 ± 1.0 | 59.3 ± 2.0 | 73.0 ± 12.3 | 92.9 ± 1.8 | 63.4 ± 0.7 | 5195 |
| AlexNet FV RF | 83.3 ± 10.7 | 54.3 ± 31.7 | 81.7 ± 0.3 | 49.3 ± 32.0 | 78.0 ± 12.0 | 73.0 ± 15.7 | 76.5 ± 5.8 | 89.3 ± 2.0 | 74.7 ± 6.7 | 88.0 ± 0.9 | 75.8 ± 0.4 | 166 |
| InceptionV3 FV RF | 40.7 ± 9.3 | 53.3 ± 11.3 | 60.3 ± 5.0 | 37.3 ± 21.7 | 27.7 ± 6.3 | 47.7 ± 13.0 | 16.5 ± 0.2 | 35.3 ± 13.3 | 32.0 ± 10.7 | 81.4 ± 4.1 | 44.6 ± 0.7 | 168 |
| ResNet18 FV RF | 63.7 ± 2.3 | 37.0 ± 5.7 | 78.7 ± 12.7 | 52.5 ± 21.1 | 73.3 ± 4.0 | 64.3 ± 5.0 | 61.8 ± 6.8 | 61.0 ± 9.7 | 69.3 ± 7.3 | 92.7 ± 2.0 | 66.4 ± 2.1 | 167 |
| AlexNet FV SVM | 93.7 ± 2.3 | 53.7 ± 16.3 | 90.7 ± 4.0 | 59.6 ± 15.2 | 77.7 ± 14.3 | 87.7 ± 9.7 | 82.8 ± 6.9 | 97.3 ± 1.3 | 81.3 ± 10.0 | 91.1 ± 2.5 | 82.4 ± 0.2 | 1541 |
| InceptionV3 FV SVM | 46.0 ± 14.0 | 45.0 ± 20.3 | 58.3 ± 3.0 | 42.2 ± 20.4 | 24.0 ± 5.3 | 43.7 ± 3.7 | 13.0 ± 1.1 | 26.7 ± 5.0 | 76.7 ± 3.7 | 41.3 ± 1.9 | 41.3 ± 1.9 | 1511 |
| ResNet18 FV SVM | 71.3 ± 11.3 | 35.3 ± 1.3 | 59.3 ± 10.3 | 51.6 ± 31.1 | 77.0 ± 8.3 | 76.7 ± 4.7 | 57.5 ± 0.5 | 73.3 ± 4.7 | 77.7 ± 9.7 | 94.5 ± 1.3 | 71.3 ± 1.5 | 1535 |