Object detection scene SSD-Mobilenetv1-FPN

Based on SSD_MobileNetV1_FPN (using MobileNetV1 network and FPN network as Backbone's SSD network).Similar to the fasterRcnn native model, there are four output tensors, which are the target number, confidence, coordinate frame and class ID.

SSD, like Yolo, uses a CNN network for detection, but uses multi-scale feature maps, and its basic architecture is shown in Figure 3.The core design concepts of SSD are summarized in the following three points:

(1) Using multi-scale feature maps for detection

(2) Detection using convolution

(3) Set a priori box

MobileNet is a network based on depth-level separable convolution. In fact, this structure first appeared in the inception of GoogleNet v3&