Table 7 Comparison of model complexity, accuracy, false positive/negative rates, and training time
From: Criminal emotion detection framework using convolutional neural network for public safety
CNN Architectures | Accuracy | False positive rate (FPR) | False negative rate (FNR) | Parameters (in millions) | Training time (min) | Inference time (ms) |
|---|---|---|---|---|---|---|
CNN | 96.4% | 0.9% | 1.2% | 1.2M | 45 | 10.23 |
ResNet-50 | 69.41% | 8.2% | 10.5% | 25.6M | 31.73 | 25 |
VGGNet | 97.5% | 2.3% | 3.1% | 138M | 90 | 18.31 |
LeNet-5 | 98.6% | 0.6% | 0.9% | 0.06M | 7.78 | 5 |
