remove conflicts

Author: Shashank Tyagi

main.py

@@ -10,7 +10,7 @@
 
 with tf.Session() as sess:
 		print 'Building Graph...'
-		net = HyperFace(sess)
+		net = HyperFace(sess,tf_record_file_path='aflw_train.tfrecords')
 		print 'Graph Built!'
 		sess.run(tf.global_variables_initializer())
 		net.print_variables()

model.py

@@ -5,7 +5,7 @@
 
 class HyperFace(object):
 
-	def __init__(self, sess):
+	def __init__(self, sess,tf_record_file_path=None):
 
 		self.sess = sess
 		self.batch_size = 2
@@ -22,6 +22,9 @@ def __init__(self, sess):
 		self.weight_pose = 5
 		self.weight_gender = 2
 
+		#tf_Record Paramters
+		self.filename_queue = tf.train.string_input_producer([tf_record_file_path], num_epochs=self.num_epochs)
+
 		self.build_network()
 
 
@@ -33,7 +36,9 @@ def build_network(self):
 		self.visibility = tf.placeholder(tf.float32, [self.batch_size,21], name='visibility')
 		self.pose = tf.placeholder(tf.float32, [self.batch_size,3], name='pose')
 		self.gender = tf.placeholder(tf.float32, [self.batch_size,2], name='gender')
-
+		
+		self.X = self.load_from_tfRecord(self.filename_queue)
+		
 		net_output = self.network(self.X) # (out_detection, out_landmarks, out_visibility, out_pose, out_gender)
 
 		loss_detection = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(net_output[0], self.detection))
@@ -57,10 +62,6 @@ def train(self):
 		writer = tf.summary.FileWriter('./logs', self.sess.graph)
 		loss_summ = tf.summary.scalar('loss', self.loss)
 
-
-
-
-
 	def network(self,inputs):
 
 		with slim.arg_scope([slim.conv2d, slim.fully_connected],
@@ -102,6 +103,32 @@ def network(self,inputs):
 
 		return [out_detection, out_landmarks, out_visibility, out_pose, out_gender]
 
+	def load_from_tfRecord(self,filename_queue):
+		
+		reader = tf.TFRecordReader()
+		_, serialized_example = reader.read(filename_queue)
+		
+		features = tf.parse_single_example(
+			serialized_example,
+			features={
+				'image_raw':tf.FixedLenFeature([], tf.string),
+				'width': tf.FixedLenFeature([], tf.int64),
+				'height': tf.FixedLenFeature([], tf.int64)
+			})
+		
+		image = tf.decode_raw(features['image_raw'], tf.float32)
+		orig_height = tf.cast(features['height'], tf.int32)
+		orig_width = tf.cast(features['width'], tf.int32)
+		
+		image_shape = tf.pack([orig_height,orig_width,3])
+		image_tf = tf.reshape(image,image_shape)
+
+		resized_image = tf.image.resize_image_with_crop_or_pad(image_tf,target_height=self.img_height,target_width=self.img_width)
+		
+		images = tf.train.shuffle_batch([resized_image],batch_size=self.batch_size,num_threads=1,capacity=50,min_after_dequeue=10)
+		
+		return images
+
 	def print_variables(self):
 		variables = slim.get_model_variables()
 		print 'Model Variables:'

with SPN/main.py

@@ -4,7 +4,7 @@
 
 weights_path = '/Users/shashank/Tensorflow/SPN/weights/'
 imgs_path = '/Users/shashank/Tensorflow/CSE252C-Hyperface/git/truth_data.npy'
-
+tf_record_file_path = 'aflw_train.tfrecords'
 if not os.path.exists('./logs'):
 	os.makedirs('./logs')
 
@@ -13,7 +13,7 @@
 
 with tf.Session() as sess:
 		print 'Building Graph...'
-		model = Network(sess)
+		model = Network(sess,tf_record_file_path=tf_record_file_path)
 		print 'Graph Built!'
 		sess.run(tf.global_variables_initializer())
 		model.train()

with SPN/model.py

@@ -7,7 +7,7 @@
 
 class Network(object):
 
-	def __init__(self, sess):
+	def __init__(self, sess,tf_record_file_path=None):
 
 		self.sess = sess
 		self.batch_size = 2
@@ -26,6 +26,8 @@ def __init__(self, sess):
 		self.weight_pose = 5
 		self.weight_gender = 2
 
+		#tf_Record Paramters
+		self.filename_queue = tf.train.string_input_producer([tf_record_file_path], num_epochs=self.num_epochs)
 		self.build_network()
 
 
@@ -39,10 +41,11 @@ def build_network(self):
 		self.gender = tf.placeholder(tf.float32, [self.batch_size,2], name='gender')
 
 
-		theta = self.localization_squeezenet(self.X)
+		
+		self.X = self.load_from_tfRecord(self.filename_queue,resize_size=(self.img_width,self.img_height))
 
+		theta = self.localization_squeezenet(self.X)
 		self.T_mat = tf.reshape(theta, [-1, 2,3])
-
 		self.cropped = transformer(self.X, self.T_mat, [self.out_height, self.out_width])
 
 		net_output = self.hyperface(self.cropped) # (out_detection, out_landmarks, out_visibility, out_pose, out_gender)
@@ -236,7 +239,31 @@ def predict(self, imgs_path):
 		brk()
 
 
+	def load_from_tfRecord(self,filename_queue,resize_size=None):
+		
+		reader = tf.TFRecordReader()
+		_, serialized_example = reader.read(filename_queue)
+		
+		features = tf.parse_single_example(
+			serialized_example,
+			features={
+				'image_raw':tf.FixedLenFeature([], tf.string),
+				'width': tf.FixedLenFeature([], tf.int64),
+				'height': tf.FixedLenFeature([], tf.int64)
+			})
+		
+		image = tf.decode_raw(features['image_raw'], tf.float32)
+		orig_height = tf.cast(features['height'], tf.int32)
+		orig_width = tf.cast(features['width'], tf.int32)
+		
+		image_shape = tf.pack([orig_height,orig_width,3])
+		image_tf = tf.reshape(image,image_shape)
 
+		resized_image = tf.image.resize_image_with_crop_or_pad(image_tf,target_height=resize_size[1],target_width=resize_size[0])
+		
+		images = tf.train.shuffle_batch([resized_image],batch_size=self.batch_size,num_threads=1,capacity=50,min_after_dequeue=10)
+		
+		return images
 
 	def load_weights(self, path):
 		variables = slim.get_model_variables()