def setup(self, index):
self.index = index
if len(weight_sets) <= self.index:
return False
s = weight_sets[self.index]["size"]
self.size = (s[0], s[1])
self.num_layers = weight_sets[self.index]["num_layers"]
self.session_name = "weight_sets/" + weight_sets[self.index]["session_name"]
self.formatter = dataformat.DataFormat(self.size[0])
self.close_down()
print(self.session_name)
num_intra_class = 10
num_inter_class = 20
self.comparator = momentnet.Comparator((2, self.size[0]), self.size[1], num_intra_class=num_intra_class, num_inter_class=num_inter_class, layers=self.num_layers)
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
self.running = True
self.sess = tf.Session(config=config)
self.sess.run(tf.global_variables_initializer())
self.comparator.load_session(self.sess, self.session_name)
return True
print("testset shapes: ", data.shape, labels.shape)
if args.template is not None:
template_path = os.path.join("templates", args.template)
else:
template_path = os.path.join("templates", "default")
templates, template_labels = dataformat.read_template_directory(
formatter, template_path, with_flip=False)
print("template shapes: ", templates.shape, template_labels.shape)
num_intra_class = 10
num_inter_class = 20
comparator = momentnet.Comparator((2, input_size[0]),
input_size[1],
num_intra_class=num_intra_class,
num_inter_class=num_inter_class,
layers=num_layers)
sess = tf.Session()
sess.run(tf.global_variables_initializer())
comparator.load_session(sess, session_name)
classes, raw = comparator.process(
sess, np.reshape(data, [-1, input_size[0] * 2]),
np.reshape(templates, [-1, input_size[0] * 2]))
print(classes)
sess.close()
print(
"Percent correct = ",
np.count_nonzero(labels[:, 0] == template_labels[classes, 0]) * 100.0 /
template_dir = "neo_large"
formatter = dataformat.DataFormat(256)
templates, template_labels, raws = dataformat.read_template_directory(
formatter,
os.path.join("templates", template_dir),
with_flip=True,
return_raw=True)
print(templates.shape)
X = np.reshape(templates, [templates.shape[0], -1])
perform_embedding = True
if perform_embedding:
comparator = momentnet.Comparator((2, 256),
32,
num_intra_class=10,
num_inter_class=20,
layers=5,
lambdas=(5, 0.5, 5))
session_name = os.path.join(os.path.dirname(__file__), '..', "weight_sets",
"neo_large_5L_flip", "test")
sess = tf.Session()
sess.run(tf.global_variables_initializer())
comparator.load_session(sess, session_name)
X = comparator.embed(sess, X)
sess.close()
#----------------------------------------------------------------------
# Scale and visualize the embedding vectors
