regarding two or multiple convolutional layers connecting with each other before a max pooling layer

[surfreta]

Hello, in accordance with the Segnet architecture, there are two convolutional layers connecting with each other befor attaching another maxpooling layer in each building block, as shown in the following figure. I am not very clear about the logic of designing this way. Why just having a larger convolution layers combining two. What are the real advantage of having several convolutional layers sequentially in each block.

[TimoSaemann]

After each conv layer follows a ReLU layer. In this way you include more non-linearities (ReLU), which gives the network more power (roughly said).