MTAL: A Novel Chinese Herbal Medicine Classification Approach with Mutual Triplet Attention Learning

Figure 3

Comparison results of different models in terms of accuracy and loss based on the same basic network. (a) Comparison of accuracy of different models, including single ResNet18 model (SigRes18), single ResNet18+triplet attention (SigAttRes18), two single ResNet18 with knowledge learned from the other ResNet18 (MulRes18-1 and MulRes18-2), and two SigAttRes18 with knowledge learned from the other SigAttRes50 (MTALRes18-1 and MTALRes18-2). (b) Comparison of accuracy of different models, including single ResNet50 model (SigRes50), single ResNet50+triplet attention (SigAttRes50), two single ResNet50 with knowledge learned from the other ResNet50 (MulRes50-1 and MulRes50-2), and two SigAttRes50 with knowledge learned from the other SigAttRes50 (MTALRes50-1 and MTALRes50-2). (c) The comparison of losses of corresponding models of (a). (d) The comparison of losses of corresponding models of (b).

(a)

(b)