Table 2.
Distribution of Classes Across the Entire Dataset
| Architecture (Ref. #) | Trainable Parameters (millions) | Loss (Lower Is Better) | % of Accuracy (Higher Is Better) |
|---|---|---|---|
| DenseNet 121 (9) | 7.0 | 0.36 | 90.8 |
| Inception V3 (6) | 21.9 | 1.06 | 79.5 |
| Resnet (7) | 23.6 | 3.24 | 44.9 |
| VVGNet 16 (5) | 14.7 | 4.33 | 4.4 |
| Xception (8) | 20.9 | 0.34 | 91.1 |
Results of stage 1, in which the 5 architectures are compared, having been trained on only three-fourths of the training data at a time. Performance of 5 network designs. Loss is a special index of inaccuracy which gives penalties for confident wrong predictions more than unconfident ones.