chapter-2/1-predict-class MobileNetV3Small

[zoldaten]

chapter ends with MobileNetV2 example. i extend a bit using MobileNetV3, as it is a bit fresher. but perfomance results shows strange things.

def predict2(img_path):
    img = image.load_img(img_path, target_size=(224, 224))
    model = tf.keras.applications.**MobileNetV2**()
    img_array = image.img_to_array(img)
    img_batch = np.expand_dims(img_array, axis=0)
    img_preprocessed = preprocess_input(img_batch)
    prediction = model.predict(img_preprocessed)
    print(decode_predictions(prediction, top=3)[0])

%timeit -r 3 predict2(IMG_PATH) 

MobileNetV2 gives 912 ms ± 5.02 ms per loop

while as MobileNetV3 gives 1.04 s ± 19.9 ms per loop from tensorflow.keras.applications import MobileNetV3Small

def predict3(img_path):
    img = image.load_img(img_path, target_size=(224, 224))
    model = tf.keras.applications.MobileNetV3Small()
    img_array = image.img_to_array(img)
    img_batch = np.expand_dims(img_array, axis=0)
    img_preprocessed = preprocess_input(img_batch)    
    prediction = model.predict(img_preprocessed)
    #print(prediction)
    print(decode_predictions(prediction, top=3)[0]) 
%timeit -r 3 predict3(IMG_PATH) 

very strange. MobileNetV3Small should be faster.