White Box Concept Of Receptive Fields

Source: Deep Learning on Medium

Hey gals/guys !! welcome all !!

It’s been long I have came up with a blog :(

Anyways today we will be painting the Black Box of Receptive Field and make it white box :) also we will also come up with FORMULA for calculating the Receptive field of the network.

This article can be used as an add on to this article https://medium.com/mlreview/a-guide-to-receptive-field-arithmetic-for-convolutional-neural-networks-e0f514068807 by Dang Ha The Hien

Please go through the image carefully as I have explained the whole concept in a crisp manner.

receptive field -by ashish johnson

Receptive Field is a relative concept and the reference will be the actual image

I hope you have read the image carefully as this image strictly forms the basis of the subsequent explanation. If not please go the through the image once again !!

Now using the above image as the reference lets go on to come up with a formula for calculating the Receptive Field of the network…. Sounds interesting na…!! :)

Consider adding THIRD CONVOLUTION LAYER with (5×5) kernel.

Please note few points for THIRD CONVOLUTION

INPUT : (6X6) image (output of the second convolution)

KERNEL : (5X5)

So far so good..!! just chill

So the output image after the third convolution will be (2×2)image.

Why (2×2) output any guess ?? Please comment !! :)

Now for calculating the Receptive field(RF)

For first convolution

RF(final) = Kernel_size

From second convolution onward

RF(final) = Kernel_size + {RF(previous)-1}

So if we are applying third convolution with the kernel size = (5×5) then the Receptive Field will be (5+{5–1}) = 9 , that is (9×9)

Note: In the whole article the stride is always 1 and there is no padding cocept.

Thanks for finding time for reading this article, I hope you liked this and looking forward for the constructive criticisms.

So Bid Adieu !!