-
Notifications
You must be signed in to change notification settings - Fork 0
/
split_and_augment_dataset.py
187 lines (162 loc) · 8.59 KB
/
split_and_augment_dataset.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
import os
import shutil
import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf
import time
import global_vars as GLOBALS
from rotate_and_crop import rotate_and_crop as rotcrop
from train_test_val_split_class import train_test_val_split_class as train_test_val_split_class
# from pipeline import initialize_hyper
def gen_augmented_dataset(i, dataset_dir_name, augmented_dir_name, seed_num):
"""
Function creates 1 set of augmented images from images in dataset_dir_name
INPUTS:
::int:: i #the augmented dataset number
::string:: dataset_dir_name #FULL path of the original dataset
::string:: augmented_dir_name #name of directory of augmented dataset, which will be created in parent folder of dataset_dir_name
::int:: seed_num #random seed number
"""
np.random.seed(seed_num) #Change the random seed to get a new set of images
root = os.path.dirname(dataset_dir_name) #directory in which dataset_dir_name is in
dataset_dir = dataset_dir_name #directory of original dataset
augmented_dir = os.path.join(root, augmented_dir_name) #directory of augmented images
if os.path.isdir(augmented_dir) == False: #if directory does not exist, make director
os.mkdir(augmented_dir)
last_img_num = 535 * i
start_time = time.time()
for entry in os.scandir(dataset_dir):
if entry.is_file() and (not entry.path.endswith(".txt")):
entry_string = os.path.basename(entry)
image = plt.imread(entry.path)
splitted = entry_string.split("_")
new_filenum = int(splitted[0]) + last_img_num #new file number
####################################################################
#Data augmentation:
angle = np.random.uniform(-15,15) #angle to rotate and crop (degrees)
new_image = rotcrop(image,angle)
if np.random.randint(2) == 1:
new_image = tf.image.flip_left_right(new_image) #randomly flip the image along vertical axis
brightnes_amount = np.random.uniform(0,0.2) #brightness amount, float between [0,1)
new_image = tf.image.adjust_brightness(new_image, brightnes_amount) #change brightness
saturation_amount = np.random.uniform(1,2) #saturation amount
new_image = tf.image.adjust_saturation(new_image, saturation_amount) #change saturation
####################################################################
tf.keras.preprocessing.image.save_img(os.path.join(augmented_dir, str(new_filenum) + "_" + splitted[1]), new_image) #save image
end_time = time.time()
print("Augmented dataset number " + str(i) + ": Augmenting all the images took", end_time-start_time, "seconds")
return True
def augment_dataset(n, dataset_dir_name):
"""
Function creates n sets of augmented images from images in dataset_dir_name
INPUTS:
::int:: n #number of sets of augmented images to create
::string:: dataset_dir_name #FULL path of the original dataset
"""
for i in range(1,n+1,1):
seed_num = i*10
gen_augmented_dataset(i, dataset_dir_name, "augmented" + str(i).zfill(2), seed_num)
return True
def merge_augmented_datasets(n, dataset_dir_name, merged_dir_name, txt_filename):
"""
This function merges the augmented datasets into one folder
INPUTS:
::int:: n #number of sets of augmented images to create
::string:: dataset_dir_name #FULL path of the original dataset
::string:: merged_dir_name #Name of the merged directory of augmented images
::string:: txt_filename #Base path (not full path) of the txt label file
"""
root = os.path.dirname(dataset_dir_name) #parent directory in which dataset_dir_name is in
ma_dir = os.path.join(root, merged_dir_name) #merged directory of augmented images
if os.path.isdir(ma_dir) == False: #if directory does not exist, make director
os.mkdir(ma_dir)
full_txt_filename = os.path.join(dataset_dir_name, txt_filename) #full filename of txt label file
merged_txt_filename = os.path.join(ma_dir, txt_filename) #path of txt file containing all the merged labels
#copy files from dataset_dir_name to to ma_dir
for base_filename in os.listdir(dataset_dir_name):
if base_filename[-3:len(base_filename)] != 'txt':
full_filename = os.path.join(dataset_dir_name, base_filename)
if os.path.isfile(full_filename):
shutil.copy(full_filename, ma_dir)
#copy files from augmented dataset to ma_dir
for i in range(1,n+1,1):
augmented_dir = os.path.join(root, "augmented" + str(i).zfill(2))
for base_filename in os.listdir(augmented_dir):
if base_filename[-3:len(base_filename)] != 'txt':
full_filename = os.path.join(augmented_dir, base_filename)
if os.path.isfile(full_filename):
shutil.copy(full_filename, ma_dir)
house_info = get_house_info_as_list(full_txt_filename)
#copy contents of the original txt label file
for i in range(0,n+1,1):
with open(merged_txt_filename, 'a') as f:
for line in house_info:
f.write(line)
return True
def get_house_info_as_list(filename):
"""
INPUT:
::string:: filename #FULL filename of txt file
OUTPUT:
::list:: housing_info_list #list containing the lines of filename
"""
housing_info_list = []
with open(filename, 'r') as f:
for line in f:
housing_info_list.append(line)
return housing_info_list
def split_and_augment_train_dataset(ratio, dataset_full_path, txt_filename_raw, n, split=True, augment=True):
"""
This function splits the original dataset and then applies data augmentation
to the train dataset. The final dataset containing augmented train images
and labels is called train_augmented
INPUTS:
::tuple of 3 floats:: ratio #train, val, test split ratio
::string:: dataset_full_path #FULL path of the original dataset
::string:: txt_filename_raw #Base path (not full path) of the txt label file
::int:: n #number of sets of augmented images to create
::boolean:: split #Whether or not to split the dataset
::boolean:: augment #Whether or not to augment the train dataset
"""
train_dir = ''
if split == True:
start_time = time.time()
# splitter = train_test_val_split_class(dataset_full_path, txt_filename_raw, train_val_test_ratio)
splitter = train_test_val_split_class(dataset_full_path, txt_filename_raw, ratio)
splitter.do_split()
end_time = time.time()
print("Splitting the dataset took " + str(end_time-start_time) + " seconds")
train_dir = splitter.train_dir
if train_dir == '':
train_dir = os.path.join(os.path.dirname(dataset_full_path), 'splitted_dataset', 'train')
if os.path.isdir(train_dir) == False:
print("ERROR: train_dir does not exist")
return False
if augment == True:
augment_dataset(n, train_dir)
start_time = time.time()
merge_augmented_datasets(n, train_dir, 'train_augmented', 'train_' + txt_filename_raw)
end_time = time.time()
print('Merging the augmented datasets took ' + str(end_time-start_time) + ' seconds')
return True
################################################################################
'''
if __name__ == '__main__':
#CHANGE THIS STUFF IF NEEDED:
config = initialize_hyper('config.yaml')
print(config)
if config is None:
print("error in initialize_hyper")
sys.exit(1)
GLOBALS.CONFIG=config
# n = 2 #number of times to augment the original train set
n = GLOBALS.CONFIG['augmentation_multiplier'] - 1
dataset_name = 'raw_dataset' #name of the dataset
train_val_test_ratio = GLOBALS.CONFIG['train_val_test_ratio']#(0.70,0.10,0.20) #train, val, test ratio
txt_filename_raw = 'HousesInfo.txt' #name of the txt label file in the original dataset
############################################################################
#It is assumed that this script is in the same directory as the raw_dataset
current_working_dir = os.getcwd() #current working directory
dataset_full_path = os.path.join(current_working_dir, dataset_name) #FULL path of the original dataset
split_and_augment_train_dataset(train_val_test_ratio, dataset_full_path, txt_filename_raw, n, split=True, augment=True)
'''