An Agile and Efficient Neural Network Based on Knowledge Distillation for Scene Text Detection

Figure 4

The pruning process of the hereby proposed method. The first convolution layer has fewer kernels, while the second has fewer convolution channels.