My training prototxt file uses images of size 400x400. I dont use any cropping, image resizing in the training. I am finetuning the Alexnet for my problem. In my deploy.protxt file I define image size as 350x350. I am using matcaffe for training as well as testing.
I expect an error message as I use different image sizes in the train.prototxs and deploy.prototxt. However I didnt get any error messages, the matlab code works fine.
Since this contains a fully connected layer, the image sizes have to be same. In that case why I didnt get any errors?
May be this is due to the strides used? 400x400 and 350x350 images will give the same size of the feature maps at the layer before the fully connected one?
