load_model.py

import tensorflow as tf
import scipy.misc
from timeit import default_timer as timer
import cv2
import numpy as np


def load_graph(frozen_graph_filename):
    # We load the protobuf file from the disk and parse it to retrieve the 
    # unserialized graph_def
    with tf.gfile.GFile(frozen_graph_filename, "rb") as f:
        graph_def = tf.GraphDef()
        graph_def.ParseFromString(f.read())

    # Then, we import the graph_def into a new Graph and returns it 
    with tf.Graph().as_default() as graph:
        # The name var will prefix every op/nodes in your graph
        # Since we load everything in a new graph, this is not needed
        tf.import_graph_def(graph_def, name='')
    return graph


graph = load_graph('test_frozen_model.pb')
sess = tf.Session(graph=graph)

im_path = 'CHANGE_THIS'

im = cv2.imread(im_path)
im = cv2.cvtColor(im, cv2.COLOR_BGR2RGB)
im = np.asarray(im)

image_pl = graph.get_tensor_by_name('Placeholder_1:0')
softmax = graph.get_tensor_by_name('Validation/decoder/Softmax:0')

output = sess.run([softmax], feed_dict={image_pl: im})
shape = im.shape
output = output[0][:, 1].reshape(shape[0], shape[1])

threshold = 0.5
im_threshold = output > threshold
im_threshold = np.uint8(255*im_threshold)

cv2.imwrite('TEST.jpg', im_threshold)