A: What receptive field?
Crown Point argument is a characteristic diagram convolutional network output pixel in the size of the mapping area on the original image, the pixel values that the white layer, wherein FIG convolution output will change in the picture as possible of the area of a pixel in which will be affected and changed, then the pixel area in the original image feature map is receptive field of the pixel.
As shown above, the original CONV1 in FIG. 2 elements in the frame area calculated to produce the blue, if the value of the blue picture frame area change occurs in the value of 2 CONV1 it will be affected. In the blue picture frame area is 3 * 3 Conv1 so receptive field size is 3 * 3.
Similarly, the value of the element is affected Conv2 picture frame is generated in blue-green, so Conv2 receptive field size is 5 * 3.