Since there may be minor differences between similar vehicles, we use a blocked approach to focus on the local characteristics of the vehicle.<br><br>To get a local area, I divided the RES-net-generated feature map, which is size 112 x 112 x 64, into two overlapping local regions. <br><br>These two areas are represented by upper-local and local, representing the upper and lower areas of the feature image, respectively.<br><br>The size of the two feature areas is 56 x 112 x 64, and they will generate a distinct block feature through the feature map to eventually enter the network.<br><br>Therefore, we obtain local key features of a vehicle image through a local network.
正在翻译中..