MobileNet Image Prediction - Keras

 

 In this post, we are gonna make a machine learning model but instead of using a self made neural network, we would use Mobilenet model which is already trained a hundred thousand times and has thousands of neurons and interconnected layers.

So we just have to download the model as code mentioned in the code snippet and we would not have to make a complex neural network and train it for many epochs. The models are uploaded on the internet and you could also use other trained models like VGG16 and many other.

In this code, I am gonna predict dog breed with Mobilenet model. You can download any picture from the internet and try to predict it. This model is no human, So it would not be that much accurate.

This model would give you top 5 probabilities of your input image and would be somewhat more accurate than your self made neural networks.

Preview:

  import matplotlib.pyplot as plt
  import numpy as np
  from numpy import random
  import tensorflow as tf
  from tensorflow.keras.applications import imagenet_utils
  from tensorflow.keras.applications.imagenet_utils import decode_predictions
  from tensorflow import keras
  from tensorflow.keras.preprocessing import image
  from tensorflow.keras.models import Sequential
  from tensorflow.keras.optimizers import Adam
  from tensorflow.keras.preprocessing.image import ImageDataGenerator
  from tensorflow.keras.metrics import categorical_crossentropy


  # download the mobilenet function 
  mobile = tf.keras.applications.mobilenet.MobileNet()

  # create a function to preprocess the image to required size and dimension
  def prepare_image(file):
     img_path = "C:/data/cats_vs_dogs/test/dog/" # select the image directory
     img = image.load_img(img_path + file, target_size=(224,224))
     img_array = image.img_to_array(img)
     img_array_expand_dims = np.expand_dims(img_array, axis=0)
     return keras.applications.mobilenet.preprocess_input(img_array_expand_dims)


  preprocessed_image = prepare_image("126.jpg") # enter the name of any image you want to predict
  predictions = mobile.predict(preprocessed_image)
  results = imagenet_utils.decode_predictions(predictions)
  print(results)