I didn't understand the process of convolution layer fully.
After convolution of input image with kernels we will be getting some feature maps now for the next convolution layer in input we are having the many feature map instead of single image. So, my question is are we need to make them a single image then proceed for the next convolution layer. If yes then how we will get the single map from all the images.
If no then how to proceed with these many feature maps.
Thank You
Dear Madhuri Panwar,
Look the links. May be useful/
http://cs231n.github.io/convolutional-networks/
http://deeplearning.net/tutorial/lenet.html
Regards, Shafagat
Multi-resolution priciple in wavelet theory may help you to understand many feature approach. Convolution is simpley only to express the reprizentation by kerlens as like the expansion by basic vectors. If you want to do something else, you use another kernel set, and so on. You could not get any valuable imformation by convolution by kernels. So, matching with something is needed. As many features in an input image, many maps are needed for get required results.
http://www.cs.toronto.edu/~hinton/absps/NatureDeepReview.pdf
This is an amazing review paper that gives an abstract explanation of how does the convolution process works throughout hierarchal layers in convolutional neural nets.
- Badges
- Science topic