Table 9.
Accuracy of architectures for image classification.
| Model | Accuracy in Animal Farming (%) | Top-1 Accuracy in ImageNet (%) | Reference |
|---|---|---|---|
| Early versions of CNN | |||
| AlexNet [33] | 60.9–97.5 | 63.3 | [55,76,131] |
| LeNet5 [27] | 68.5–97.6 | [156] | |
| Inception family | |||
| Inception V1/GoogLeNet [35] | 96.3–99.4 | [66,76] | |
| Inception V3 [158] | 92.0–97.9 | 78.8 | [63,76,120] |
| Inception ResNet V2 [160] | 98.3–99.2 | 80.1 | [69,120] |
| Xception [162] | 96.9 | 79.0 | [120] |
| MobileNet family | |||
| MobileNet [149] | 98.3 | [120] | |
| MobileNet V2 [164] | 78.7 | 74.7 | [120] |
| NASNet family | |||
| NASNet Mobile [150] | 85.7 | 82.7 | [120] |
| NASNet Large [150] | 99.2 | [120] | |
| Shortcut connection networks | |||
| DenseNet121 [39] | 75.4–85.2 | 75.0 | [120,151] |
| DenseNet169 [39] | 93.5 | 76.2 | [120] |
| DenseNet201 [39] | 93.5–99.7 | 77.9 | [69,76,120] |
| ResNet50 [36] | 85.4–99.6 | 78.3 | [69,76,151], etc. |
| ResNet101 [36] | 98.3 | 78.3 | [120] |
| ResNet152 [36] | 96.7 | 78.9 | [120] |
| VGGNet family | |||
| VGG16 [34] | 91.0–100 | 74.4 | [49,107,151], etc. |
| VGG19 [34] | 65.2–97.3 | 74.5 | [120,131] |
| YOLO family | |||
| YOLO [148] | 98.4 | [74] | |
| DarkNet19 [171] | 95.7 | [76] | |
Note: “Net” in model names is network, and number in model names is number of layers of network. AlexNet is network designed by Alex Krizhevsky; CNN is convolutional neural network; DenseNet is densely connected convolutional network; GoogLeNet is network designed by Google Company; LeNet is network designed by Yann LeCun; NASNet is neural architecture search network; ResNet is residual network; VGG is visual geometry group; Xception is extreme inception network; and YOLO is you only look once. “” indicates missing information.