Created by: shippingwang

TODO:

• Check image classification pre-trained models one more time, to make sure it's consistent with the architecture, such as:

why don't VGG pre-trained model have offset params meanwhile vgg.py support bias? Need to update the code

• some pre-trained models compressed on the mac, and will obtain redundancy folder: __MACOSX when decompressing on the Linux machine.
• Unify filename, format: #2100 (closed)
• Bug: googlenet #2187 (closed)
• Bug: resnet34 #2099