def create_model(self, brand=None):
layers = tf.keras.layers
dropout_rate = 0.2
model = tf.keras.models.Sequential()
model.add(
layers.Conv2D(64, (4, 4),
activation='relu',
input_shape=(self.imageSize[0], self.imageSize[1],
1)))
model.add(layers.MaxPooling2D((3, 3)))
# model.add(layers.BatchNormalization())
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.Flatten())
model.add(layers.Dense(128, activation='relu'))
model.add(layers.Dropout(dropout_rate))
num_labels = len(self.labels) if brand is None else len(
self.brand_shoes_map[detailing.label_brand(brand)])
model.add(layers.Dense(num_labels, activation='softmax'))
model.summary()
model.compile(optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
return model
def create_testing_data(self):
for query in self.labels:
brand = detailing.label_brand(query)
path = self.test_data_dir + query
for filename in listdir(path):
self.test_data.append(self.create_np_info(filename, path, brand))
np.random.shuffle(self.test_data)
np.save("./npy_data/test_whole_data.npy", self.test_data)
def create_testing_data(self):
for query in self.shoe_names:
brand_label = brand_detailing.label_brand(query)
path = self.test_data_dir + query
for filename in listdir(path):
info = self.create_np_info(filename, path, brand_label)
if info is None:
continue
else:
self.test_data.append(info)
np.random.shuffle(self.test_data)
np.save("./npy_data/test_whole_data.npy", self.test_data)
def run_all_networks(self):
# copy to avoid unwanted mutations in original data-set
all_train_data = self.train_data.copy()
all_test_data = self.test_data.copy()
brand_results = {}
# create folder for storing all brand network states
network_paths = "./network_states"
try:
os.mkdir(network_paths)
except OSError:
print()
# get all data under one brand, execute network for that data
for brand in detailing.known_brands:
if brand != "nike":
continue
if detailing.label_brand(brand) not in self.brand_shoes_map.keys():
continue
train_data = []
test_data = []
print(brand)
for i in range(len(all_train_data)):
train = all_train_data[i]
if detailing.known_brands[train[2]] == brand:
train_data.append(train)
for i in range(len(all_test_data)):
test = all_test_data[i]
if detailing.known_brands[test[2]] == brand:
test_data.append(test)
np.random.shuffle(train_data)
np.random.shuffle(test_data)
test_loss, test_acc = self.run_network(brand, train_data,
test_data, network_paths)
brand_results.update({brand: [test_loss, test_acc]})
for brand in brand_results.keys():
print("BRAND: " + brand)
print("Test loss: " + str(brand_results[brand][0]))
print("Test accuracy: " + str(brand_results[brand][1]) + "\n")
def create_brand_shoes_map(labels):
brand_shoes_map = {}
for model in labels:
model_brand = detailing.label_brand(model)
if model_brand == -1:
print("ERROR: " + model + " (NO KNOWN BRAND IDENTIFIER)")
continue
if model_brand not in brand_shoes_map.keys():
brand_shoes_map.update({model_brand: [model]})
else:
mbt = brand_shoes_map[model_brand]
mbt.append(model)
brand_shoes_map.update({model_brand: mbt})
print(brand_shoes_map)
with open(brand_shoe_map_path, "wb") as map_file:
pickle.dump(brand_shoes_map, map_file)
return brand_shoes_map
