yolo_net.py:

输入图像尺寸为448，cell_size 为7

self.boundary1 = self.cell_size * self.cell_size * self.num_classes  # 7 x 7 x 20

self.boundary2 = self.boundary1 + self.cell_size * self.cell_size * self.boxes_per_cell  

                                                                                                         # 7 x 7 x 20 + 7 x 7 x 2

self.output_size = (self.cell_size * self.cell_size) * (self.num_classes + self.boxes_per_cell * 5)

网络输出predict 为 fc_32层，Labels shape = [batch_size, 7, 7, 25]

类别预测:  #[batch_size, 7, 7, 20]

       predict_classes = [self.batch_size, self.cell_size, self.cell_size, self.num_class])

定位预测: #[batch_size, 7, 7, 2]

       predict_scales = [self.batch_size, self.cell_size, self.cell_size, self.boxes_per_cell]

box大小预测: #[batch_size, 7, 7, 2, 4]

        predict_boxes = [self.batch_size, self.cell_size, self.cell_size, self.boxes_per_cell, 4]

Label：

类别结果：

response = [self.batch_size, self.cell_size, self.cell_size, 1] # [batch_size, 7, 7, 1]

定位结果:

boxes= [self.batch_size, self.cell_size, self.cell_size, 1, 4]) # [batch_size, 7, 7, 1, 4]

box大小结果：

boxes=tf.tile(boxes, [1, 1, 1, self.boxes_per_cell, 1]) / self.image_size  # [batch_size, 7, 7, 2, 4]