flash1803

> @jeong-tae I think step 2 is something like: > > 1. Initialize the network with VGG pre-trained from ImageNet. > 2. Forward propagation the images and get the feature...