Input data for a convolutional neural network
I can't figure out how to convert images to the input of the convolutional network. For example, I want to recognize circles in a 1000x1000 pxs photo. The network is trained on 100x100 pxs data. I can't put one situation in my head when the circle in the input drawing is much larger (or smaller) than 100x100 pxs, as in this case, the convolutional network determines that circle if it was learned from data of a different size. I searched a lot on the Web, but my head is confused. For clarity on the entrance I want to feed the 454x430 pxs image to the network:
The network is trained on 100x100 pxs data:
At the output, I want to recognize all the circles:
The drawings are approximate.
1 answers
You need to compress the photo to 100x100 px(If the vector is increased, the system will not accept it)