Figure 4
From: Deep learning-enabled mobile application for efficient and robust herb image recognition
An illustration of the pipeline for the network compression algorithm. (a) Firstly, the large and small DNN are pre-trained on the ImageNet dataset and further trained on the herb image dataset, and they are trained individually; (b) Secondly, the recognition ability of the large DNN is transferred to the small DNN, as shown by the dotted red line; (c) Finally, network cutting is processed in a top-down way from the top layer to the bottom layer; (d) The network cutting for a single layer, wherein some unimportant filters are removed.
