pythontensorflowkerasmnist

Using pretrained models for mnist dataset


The problem when I want to use pre-trained VGG16 is that is expects shape=(None, 224, 224, 3), but found shape=(32, 28, 28). What can I do in order to use the model? or should I not use convnets for images under 244 x 244 pixels?


Solution

  • resize using tensorflow as follows:

    (x_train, y_train), (_, _) = tf.keras.datasets.mnist.load_data()
    
    print(x_trian.shape) # (60000, 28, 28)
    
    # train set / data 
    x_train = np.expand_dims(x_train, axis=-1)
    x_train = tf.image.resize(x_train, [224,224]) 
    
    print(x_train.shape) # (60000, 224, 224, 1)