Some question about Alexnet size

[Teoge]

I am new in caffe and deep learning.

When I setup caffe to train Alexnet, I found out that the tutorials of caffe instruct me to resize the image to size 256x256 before training.

But later I notice that the input size of Alenet is 227x227x3.

And then I went to read the orginal paper of Alexnet, it claims that it uses size 256x256, but it is because it randomly crops the image to achieve data augmetation.

But I searched most of the process of caffe training, and found no evidence of image crop. But since I'm new in caffe, maybe I ignore somthing.

So, my question is: what exactly is the size of input image while training Alexnet using caffe and why?

Hope someone may answer my question. Thank you.

[Hieu Do Trung]

In train_val protxt file, you can see this: transform_param {

mirror: true

crop_size: 227

mean_file: "data/ilsvrc12/imagenet_mean.binaryproto"

}

which is the implementation of "randomly crops the image to achieve data augmetation".

[Teoge]

Oh, I see it now! Thank you very much for pointing out my ignorance! I didn't fully understand the parameter in prototxt until you told me. I'll read the protocol again carefully. Many thanks!