Combine CNN with LSTM

  keras, neural-network, python, tensorflow, tflearn

I’m looking to implement a RNN along with a CNN in order to make a prediction based on two images instead of one alone with a CNN.
I’m trying to modify the alexnet model code:

def alexnet(width, height, lr, output=3):
    network = input_data(shape=[None, width, height, 1], name='input')
    network = conv_2d(network, 96, 11, strides=4, activation='relu')
    network = max_pool_2d(network, 3, strides=2)
    network = local_response_normalization(network)
    network = conv_2d(network, 256, 5, activation='relu')
    network = max_pool_2d(network, 3, strides=2)
    network = local_response_normalization(network)
    network = conv_2d(network, 384, 3, activation='relu')
    network = conv_2d(network, 384, 3, activation='relu')
    network = conv_2d(network, 256, 3, activation='relu')
    network = max_pool_2d(network, 3, strides=2)
    network = local_response_normalization(network)
    network = fully_connected(network, 4096, activation='tanh')
    network = dropout(network, 0.5)
    network = fully_connected(network, 4096, activation='tanh')
    network = dropout(network, 0.5)
    network = fully_connected(network, output, activation='softmax')
    network = regression(network, optimizer='momentum',
                         loss='categorical_crossentropy',
                         learning_rate=lr, name='targets')

    model = tflearn.DNN(network, checkpoint_path='model_alexnet',
                        max_checkpoints=1, tensorboard_verbose=0, tensorboard_dir='log')

    return model

I have my images in a np array where each element is the pixel data for one image. I’m having trouble implementing the functionality of using two images with the RNN.

I’ve seen the reshape and lstm methods of tflearn which I believe should be placed before the final fully connected layer but not sure how to specify the number of images to use.

Also, would this be easier to implement with Keras?

Source: Python Questions

LEAVE A COMMENT