The Input image size is (28,28).

Context Aggregation Networks (CAN) (25 points) We can also implement a simple CNN model with dilated convolutions. We can have a model with a large receptive field with the same spatial resolution in the hidden layers. See the network architecture in the Table 1. For your information, the dilated residual network 2 is a similar model for image classification. Table 1: The CAN model. The average pooling is a global average pooling that outputs a 10-dimensional vector 4.LReLU represents Leaky ReLU