Figure 7
From: A study on Shine-Muscat grape detection at maturity based on deep learning
Structure of SPPNet, PANet and depth separable modules. Note CBS denotes the Conv + BN + SiLu structure, where SiLU is the combined Sigmoid and ReLU activation function; UpSampling means up-sampling operation, which increases the scale of the feature map by nearest interpolation in the width and height directions; DownSampling means down-sampling operation, which compresses the height and width of the feature map by depth-wise convolution with stride = 2; Depthwise Conv (3 × 3) means depth-wise convolution operation. Conv (1 × 1) indicates the number of channels adjusted by ordinary convolution, Same as below.
