def graph_eval(input_graph_def, input_node, output_node):
#Reading images and segmentation labels
x_test, y_test, img_file, seg_file = cnn.get_images_and_labels(
IMG_TEST_DIR, SEG_TEST_DIR, cfg.NUM_CLASSES, cfg.WIDTH, cfg.HEIGHT)
# load graph
tf.import_graph_def(input_graph_def, name='')
# Get input & output tensors
x = tf.get_default_graph().get_tensor_by_name(input_node + ':0')
y = tf.get_default_graph().get_tensor_by_name(output_node + ':0')
# Create the Computational graph
with tf.Session() as sess:
sess.run(tf.initializers.global_variables())
feed_dict = {x: x_test} #, labels: y_test}
y_pred = sess.run(y, feed_dict)
# Calculate intersection over union for each segmentation class
y_predi = np.argmax(y_pred, axis=3)
y_testi = np.argmax(y_test, axis=3)
print(y_testi.shape, y_predi.shape)
cnn.IoU(y_testi, y_predi)
#cfg.ShowSegmentedImages(100, x_test, y_testi, y_predi, cfg.NUM_CLASSES) # steps by 20
print('FINISHED!')
return x_test, y_testi, y_predi, img_file, seg_file
def graph_eval(input_graph_def, input_node, output_node):
#Reading images and segmentation labels
x_test, y_test, img_file, seg_file= cnn.get_images_and_labels(IMG_TEST_DIR, SEG_TEST_DIR,
cfg.NUM_CLASSES, cfg.WIDTH, cfg.HEIGHT)
y_pred2 = np.zeros((y_test.shape))
y_pred2i = np.zeros((y_test.shape[0],y_test.shape[1],y_test.shape[2]))
# load graph
tf.import_graph_def(input_graph_def,name = '')
# Get input & output tensors
x = tf.compat.v1.get_default_graph().get_tensor_by_name(input_node+':0')
y = tf.compat.v1.get_default_graph().get_tensor_by_name(output_node+':0')
# Create the Computational graph
with tf.compat.v1.Session() as sess:
sess.run(tf.compat.v1.initializers.global_variables())
feed_dict={x: x_test}
#y_pred = sess.run(y, feed_dict) # original code
logits = sess.run(y, feed_dict) # new code
pred_DB = softmax_predict(logits) #new code
y_pred = pred_DB.eval() #new code
##alternative way to compute softmax in C++ style
#for i in range(logits.shape[0]):
# y_pred2[i]= cpp_softmax(logits[i], 224, 224, 12)
# Calculate intersection over union for each segmentation class
y_predi = np.argmax(y_pred, axis=3)
y_testi = np.argmax(y_test, axis=3)
print(y_testi.shape,y_predi.shape)
cnn.IoU(y_testi,y_predi)
##just to check the cpp_softmax works fine
#y_pred2i = np.argmax(y_pred2, axis=3)
#print(y_testi.shape,y_pred2i.shape)
#cfg.IoU(y_testi,y_pred2i)
print ('FINISHED!')
return x_test, y_testi, y_predi, y_pred2i, img_file, seg_file