save_model.py

import commentjson
import os
import sys

import collections

import numpy as np
import scipy as scp
import scipy.misc
import tensorflow as tf

sys.path.insert(1,'incl')

try:
    # Check whether setup was done correctly

    import tensorvision.utils as tv_utils
    import tensorvision.core as core
    import tensorvision.train
    import tensorflow_fcn.utils
except ImportError:
    # You forgot to initialize submodules
    logging.error("Could not import the submodules.")
    logging.error("Please execute:"
                  "'git submodule update --init --recursive'")
    exit(1)



# load the network from the working directory
# for some reason I can't get this to work without
# using tensorvision - TODO figure this out so we can
# use any checkpoint from the run directory
logdir = 'CHANGE_THIS'
image_pl = tf.placeholder(tf.float32)

hypes = tv_utils.load_hypes_from_logdir(logdir)
modules = tv_utils.load_modules_from_logdir(logdir)

image_pl = tf.placeholder(tf.float32)
image = tf.expand_dims(image_pl, 0)
pred = core.build_inference_graph(hypes, modules,
				     image=image)

# load the weights
sess = tf.Session()
saver = tf.train.Saver()
core.load_weights(logdir, sess, saver)

# freeze the graph
frozen_graph_def = tf.graph_util.convert_variables_to_constants(
        sess, sess.graph.as_graph_def(), ['Validation/decoder/Softmax'])

# remove training only nodes
frozen_graph_def = tf.graph_util.remove_training_nodes(frozen_graph_def)

# save the model
with tf.gfile.GFile('test_frozen_model.pb', 'wb') as f:
    f.write(frozen_graph_def.SerializeToString())