Error When Checking Input: Expected Flatten_input To Have 3 Dimensions, But Got Array With Shape (none, 100, 100, 1)

Using TensorFlow/Keras, I want to classify pictures into two classes, selfie and non-selfie. I have gathered samples into two filesystem folders, one for each category. I implement

Solution 1:

1) The images have one channel so this must be reflected in the input shape argument:

keras.layers.Flatten(input_shape=(100, 100, 1))

2) To load the files with tf.data API, you need to first fetch the image filenames and their corresponding labels:

image_paths, lbls = ["selfies-data/1", "selfies-data/2"], [0., 1.]

labels = []
file_names = []
for d, l in zip(image_paths, lbls):
    # get the list all the images file names
    name = [os.path.join(d,f) for f in os.listdir(d)]
    file_names.extend(name)
    labels.extend([l] * len(name))

file_names = tf.convert_to_tensor(file_names, dtype=tf.string)
labels = tf.convert_to_tensor(labels)

dataset = tf.data.Dataset.from_tensor_slices((file_names, labels))

# the rest is the same 

You may also need to expand the dimension of labels to make it have a shape of (?, 1) (instead of (?,)). To do so, you can put the following line in map_fn function:

labels = tf.expand_dims(labels, axis=-1)

3) If you have two classes, then why the last layer has 10 units? It is a binary classification problem, so make the last layer have one unit with sigmoid activation. Finally, change the loss to binary_crossentropy:

# ... 
       keras.layers.Dense(1, activation=tf.nn.sigmoid)
])

model.compile(optimizer=tf.train.AdamOptimizer(),
              loss='binary_crossentropy',
              metrics=['accuracy'])