def predict_batch(self,images,type_class):
images_predictions = tf.TensorArray(tf.float32,size=0,dynamic_size=True)
ys = tf.TensorArray(tf.float32,size=0,dynamic_size=True)
matched_images = tf.TensorArray(tf.float32,size=0,dynamic_size=True)
index = 0
basis = tf.convert_to_tensor([0,1],dtype=tf.float32)
for i in tf.range(len(images)):
gen_image = data_access.normalize(data_access.de_standardize(images[i]))
img = tf.expand_dims(gen_image,axis=0)
c = self.classifier(img)
if(self.num_classes == 2):
x = tf.subtract(c,basis)
w_list = tf.abs(x)
else:
w_list = c
w_list = tf.reshape(w_list,(w_list.shape[1],))
images_predictions = images_predictions.write(i,w_list)
y_list = tf.convert_to_tensor(type_class,dtype=tf.float32)
ys = ys.write(i,y_list)
if(tf.reduce_all(tf.equal(w_list,y_list))):
matched_images = matched_images.write(index,images[i])
index +=1
return images_predictions.stack(), ys.stack(),matched_images.stack()
def predict_batch(self,images,type_class):
"""
Classify each image received and prepare for loss function
:param images: - images tensors
:param type_class: - class chosen to influence generator. Its must be a number
"""
images_predictions = tf.TensorArray(tf.float32,size=0,dynamic_size=True)
ys = tf.TensorArray(tf.float32,size=0,dynamic_size=True)
matched_images = tf.TensorArray(tf.float32,size=0,dynamic_size=True)
index = 0
basis = tf.convert_to_tensor([0,1],dtype=tf.float32)
for i in tf.range(len(images)):
gen_image = data_access.normalize(data_access.de_standardize(images[i]))
img = tf.expand_dims(gen_image,axis=0)
c = self.classifier(img)
if(self.num_classes == 2):
x = tf.subtract(c,basis)
w_list = tf.abs(x)
else:
w_list = c
w_list = tf.reshape(w_list,(w_list.shape[1],))
images_predictions = images_predictions.write(i,w_list)
y_list = tf.one_hot(type_class,self.num_classes)
ys = ys.write(i,y_list)
if(tf.reduce_all(tf.equal(w_list,y_list))):
matched_images = matched_images.write(index,images[i])
index +=1
return images_predictions.stack(), ys.stack(),matched_images.stack()
def generate_images(self,number_of_samples,directory):
seed = tf.random.normal([number_of_samples, self.random_noise_size])
images = self.generator(seed)
if self.classifier is not None:
predictions = self.classifier(data_access.normalize(data_access.de_standardize(images)))
data_access.produce_generate_figure(directory,images,predictions,class_names)
else:
data_access.store_images_seed(directory,images,'gen_images','gan')
def predict_batch(self, images, type_class):
images_predictions = tf.TensorArray(tf.float32,
size=10,
dynamic_size=True)
ys = tf.TensorArray(tf.float32, size=10, dynamic_size=True)
matched_images = tf.TensorArray(tf.float32, size=0, dynamic_size=True)
index = 0
for i in tf.range(len(images)):
gen_image = data_access.normalize(
data_access.de_standardize(images[i]))
img = tf.expand_dims(gen_image, axis=0)
c_type = self.classifier_m.predict_image(img)
w_list = tf.one_hot(c_type, self.num_classes)
w_list = tf.reshape(w_list, (w_list.shape[1], ))
images_predictions = images_predictions.write(i, w_list)
y_list = tf.one_hot(type_class, self.num_classes)
ys = ys.write(i, y_list)
if (tf.reduce_all(tf.equal(w_list, y_list))):
matched_images = matched_images.write(index, images[i])
index += 1
return images_predictions.stack(), ys.stack(), matched_images.stack()
def generate_images(self, number_of_samples, directory):
seed = tf.random.normal([number_of_samples, 100])
images = self.generator(seed)
predictions = self.classifier_m.predict_image_vector(
data_access.normalize(data_access.de_standardize(images)))
data_access.produce_generate_figure('imgs', images, predictions)