Convolution filter returns image with same size as input

[Nerfling]

Hello,

First of all, I'm not very familiar with convolutional neural networks yet.

I tried to apply a convolution filter in this way:

Code: Select all

Convolution _convolution_filter = new Convolution(_kernel_weights); //3x3 random weight kernel
Bitmap _result = _convolution_filter.Apply(_input_bitmap);


It seems that it works but the filter returns a bitmap with the same dimensions as the input bitmap.

I've heard that through convolution the image should get smaller by a few pixels each time the filter is applied but this doesn't seem the case.

Am I doing something wrong?

Any help and advice is appreciated,
Nerfling.

[andrew.kirillov]

Hello,

First of all, I'm not very familiar with convolutional neural networks yet.

What you are doing (and asking about) has zero relation to neural networks.

I've heard that through convolution the image should get smaller by a few pixels each time the filter is applied but this doesn't seem the case.

It really depends on implementation. Images do not have to get smaller in size.