tensorflow-vgg-1's Introduction

Tensorflow VGG

This is a Tensorflow implemention of VGG forked from tensorflow-vgg repo.

The main change from the original repo are:

The Vgg class won't load the VGG model in the constructor so that you are allowed to share the model among multiple Vgg instances.
The Vgg class is now able to do training and prediction and you could shared a model among multiple Vgg instance. That makes you more flexible to design the algorithm infrastructure.

To use the VGG networks, you should download the *.npy files from VGG16 or VGG19 and put them under the models directory.

Usage

It is still under development, please stay tuned. This is a rough example for the prediction.

###Use It for Predicting

First, load the model (around 500MB) to the memory.

model = np.load("/path/to/your/vgg19.npy").item()
print("The VGG model is loaded.")

New the Vgg19 instance in the scope of your tf.Graph. Run the graph in a tf.Session.

# Design the graph.
graph = tf.Graph()
with graph.as_default():
    nn = Vgg19(model=model)

# Run the graph in the session.
with tf.Session(graph=graph) as sess:
    tf.initialize_all_variables().run()
    print("Tensorflow initialized all variables.")

    preds = sess.run(nn.preds,
                     feed_dict={
                         nn.inputRGB: imgs
                     })

You could also add a tf.train.SummaryWriter to summarize whatever you want. Then enter the command, tensorboard --log=/path/to/your/log/dir, to see the summary in the web page.

# Run the graph in the session.
with tf.Session(graph=graph) as sess:
    tf.initialize_all_variables().run()
    print("Tensorflow initialized all variables.")

    # The OP to write logs to Tensorboard.
    summary_writer = tf.train.SummaryWriter("/path/to/your/log/dir",
                                            graph=sess.graph)

    preds = sess.run(nn.preds,
                     feed_dict={
                         nn.inputRGB: imgs
                     })

Recommend Projects