def convert(save_path=FLAGS.save_path,
model_name='facenet.tfmodel',
RESULTS_NODE_NAME='embeddings'):
with tf.Graph().as_default():
with tf.Session() as sess:
saver = import_graph_def()
load_weights(sess=sess, saver=saver)
show_graph(sess.graph)
#white_list = del_moving_average_op(sess.graph)
constant_graph = graph_util.convert_variables_to_constants(
sess, sess.graph_def, [RESULTS_NODE_NAME])
#write_proto(os.path.join(save_path, model_name), constant_graph)
print type(constant_graph)
tf.train.write_graph(constant_graph,
save_path,
model_name,
as_text=False)
def convert1(save_path=FLAGS.save_path,
model_name='facenet.tfmodel',
RESULTS_NODE_NAME='embeddings'):
with tf.Graph().as_default():
graph_def = tf.GraphDef()
with open('nn4.pbtxt', 'rb') as graph_file:
graph_def.ParseFromString(graph_file.read())
tf.import_graph_def(graph_def)
with tf.Session() as sess:
#saver = import_graph_def()
saver = tf.train.Saver(tf.trainable_variables())
load_weights(sess=sess, saver=saver)
show_graph(sess.graph)
#white_list = del_moving_average_op(sess.graph)
constant_graph = graph_util.convert_variables_to_constants(
sess, sess.graph_def, [RESULTS_NODE_NAME])
#write_proto(os.path.join(save_path, model_name), constant_graph)
print type(constant_graph)
tf.train.write_graph(constant_graph,
save_path,
model_name,
as_text=False)
sess = tf.Session()
sess.run(tf.global_variables_initializer())
history = []
iterep = 500
for i in range(iterep * 30):
idx = np.random.choice(np.arange(len(y_train)), 100, replace=False)
x_batch = X_train[idx]
y_batch = y_train[idx]
sess.run(train_step,
feed_dict={x: x_batch,
y: y_batch,
phase: True})
if (i + 1) % iterep == 0:
epoch = (i + 1)/iterep
tr = sess.run([loss, accuracy],
feed_dict={x: X_train,
y: y_train,
phase: False})
t = sess.run([loss, accuracy],
feed_dict={x: X_eval,
y: y_eval,
phase: False})
history += [[epoch] + tr + t]
print history[-1]
sys.path.append('/home/abdullah/PycharmProjects/Pattern') # point to your tensorflow dir
show_graph(tf.get_default_graph().as_graph_def())
def show_graph(self):
graph = tf.get_default_graph()
show_graph(graph)
update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
with tf.control_dependencies(update_ops):
self.train_step = tf.train.AdamOptimizer(
learning_rate=self.lr).minimize(self.loss,
global_step=self.global_step)
# get accuracy
prediction = tf.argmax(predictions, 1)
equality = tf.equal(prediction, tf.argmax(self.targets, 1))
self.accuracy = tf.reduce_mean(tf.cast(equality, tf.float32))
# define summaries
tf.summary.scalar('loss_ce', ce_term)
tf.summary.scalar('loss_activacion_mean', self.mean_act_loss_term)
tf.summary.scalar('loss_total', self.loss)
if __name__ == "__main__":
from datasets.quickdraw_dataset import QuickDraw_Dataset
with tf.Session().as_default() as sess:
#t = QuickDraw_Dataset(1, 60,data_folder='./temp/quickdraw_expanded_images')
#t = Xray_dataset(1, 10, data_folder='./temp/dataset_xray_imgs')
t = Imagenet_Dataset(1, 55, data_folder='temp/imagenet_subset')
with imagenet_classifier_cam_loss_V2(t, debug=False) as model:
#model.train(save_model=False, special_t=1)
show_graph(model.graph)
#model.eval('test')
i = np.random.randint(0, len(train_images))
print('displaying image {} with class {}'.format(i, train_labels[i]))
plt.imshow(train_images[i], cmap='gray')
plt.show()
g = tf.Graph()
plt.show()
flatten_shape = np.prod(train_images.shape[1:])
with g.as_default():
X = tf.placeholder(tf.float32, [None, flatten_shape], name='X')
y = tf.placeholder(tf.float32, [None, 10], name='y')
utils.show_graph(g)
with g.as_default():
# define the model
l1 = dense(X, 32, 'h1', activation=tf.nn.sigmoid)
l2 = dense(l1, 64, 'h2', activation=tf.nn.relu)
logits = dense(l1, 10, 'out', activation=None)
# define the loss function
loss_op = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits_v2(logits=logits, labels=y))
# define the optimizer
optmizer = tf.train.RMSPropOptimizer(learning_rate=0.01)
# train operation
predictions, acts = get_slim_arch_bn(self.input_l, phase,
self.dataset.shape_target[0])
# Configure values for visualization
self.last_conv = acts['vgg_16/conv5/conv5_3']
self.softmax_weights = r"vgg_16/softmax_logits/weights:0"
self.pred = tf.nn.softmax(predictions, name='prediction')
self.loss = tf.losses.softmax_cross_entropy(self.targets, predictions)
update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
with tf.control_dependencies(update_ops):
self.train_step = tf.train.AdamOptimizer(
learning_rate=self.lr).minimize(self.loss)
# get accuracy
prediction = tf.argmax(predictions, 1)
equality = tf.equal(prediction, tf.argmax(self.targets, 1))
self.accuracy = tf.reduce_mean(tf.cast(equality, tf.float32))
if __name__ == '__main__':
dataset = Imagenet_Dataset(2, 40)
with vgg_16_batchnorm(dataset, debug=False) as model:
show_graph(tf.get_default_graph())
save_graph_txt(tf.get_default_graph())
graph = tf.get_default_graph()
writer = tf.summary.FileWriter(logdir='logdir', graph=graph)
writer.flush()