Enable MobileNet and V2 for `channels_first`

[taehoonlee]

This PR enables MobileNet and V2 for channels_first.

[taehoonlee]

Test codes:

import numpy as np
import tensornets as nets

from keras import backend as K
from keras.applications.mobilenet import MobileNet
from keras.applications.mobilenet import preprocess_input, decode_predictions

img = nets.utils.load_img('cat.png', target_size=256, crop_size=224)

model = MobileNet(weights='imagenet')
preds = model.predict(preprocess_input(img))

print(decode_predictions(preds, top=3)[0])

K.set_image_data_format('channels_first')

model2 = MobileNet(weights='imagenet')
preds = model2.predict(np.transpose(img, (0, 3, 1, 2)))

print(decode_predictions(preds, top=3)[0])

The results:

• MobileNet
  • last: [('n02124075', 'Egyptian_cat', 0.47201723), ('n03482405', 'hamper', 0.10415065), ('n02123045', 'tabby', 0.09730302)]
  • first: [('n02124075', 'Egyptian_cat', 0.472019), ('n03482405', 'hamper', 0.104150146), ('n02123045', 'tabby', 0.09730264)]
• MobileNetV2
  • last: [('n02124075', 'Egyptian_cat', 0.16725165), ('n02123045', 'tabby', 0.12793817), ('n03482405', 'hamper', 0.12389358)]
  • first: [('n02124075', 'Egyptian_cat', 0.16725159), ('n02123045', 'tabby', 0.127938), ('n03482405', 'hamper', 0.12389337)]
• Xception
  • last: [('n02123045', 'tabby', 0.17753284), ('n02124075', 'Egyptian_cat', 0.17632152), ('n03482405', 'hamper', 0.09113098)]
  • first: [('n02123045', 'tabby', 0.17753297), ('n02124075', 'Egyptian_cat', 0.17632148), ('n03482405', 'hamper', 0.09113101)]