AugmentImagesAndXML/CreateAugmentedData.py at master · deepdigger/AugmentImagesAndXML · GitHub

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
"""
Created on Tue Jun 25 17:38:08 2019
@author: bhuwan

Website: https://medium.com/@bhuwanbhattarai/image-data-augmentation-and-parsing-into-an-xml-file-in-pascal-voc-format-for-object-detection-4cca3d24b33b
"""

import cv2
import numpy as np
import os
import xml.etree.ElementTree as ET
import StaticMethods as sm

xml_input_folder_path = "AutoAugmentXML"
picture_input_path = "AutoAugmentImages"
textfile_out_path = "AugmentedXmlDataPath.txt"
augmented_img_path = "AugmentedImages"

sm.createFolder(augmented_img_path)


def pca_color_augmentation(image):
    assert image.ndim == 3 and image.shape[2] == 3
    assert image.dtype == np.uint8
    img = image.reshape(-1, 3).astype(np.float32)
    sf = np.sqrt(3.0 / np.sum(np.var(img, axis=0)))
    img = (img - np.mean(img, axis=0)) * sf
    cov = np.cov(img, rowvar=False)  # calculate the covariance matrix
    value, p = np.linalg.eig(cov)  # calculation of eigen vector and eigen value
    rand = np.random.randn(3) * 0.08
    delta = np.dot(p, rand * value)
    delta = (delta * 255.0).astype(np.int32)[np.newaxis, np.newaxis, :]
    img_out = np.clip(image + delta, 0, 255).astype(np.uint8)
    return img_out


xml_paths = [os.path.join(xml_input_folder_path, s) for s in os.listdir(xml_input_folder_path)]
# print(xml_paths)

pwd_lines = []
for xml_file in xml_paths:
    et = ET.parse(xml_file)
    element = et.getroot()
    element_objs = element.findall('object')
    element_filename = element.find('filename').text
    base_filename = os.path.join(picture_input_path, element_filename)
    print("Create augmention for file: %s" % base_filename)
    img = cv2.imread(base_filename)
    rows, cols = img.shape[:2]

    img_split = element_filename.strip().split('.jpg')

    for element_obj in element_objs:
        class_name = element_obj.find('name').text  # return name tag ie class of disease from xml file

        obj_bbox = element_obj.find('bndbox')
        # print(obj_bbox)
        x1 = int(round(float(obj_bbox.find('xmin').text)))
        y1 = int(round(float(obj_bbox.find('ymin').text)))
        x2 = int(round(float(obj_bbox.find('xmax').text)))
        y2 = int(round(float(obj_bbox.find('ymax').text)))
        # if you specify range(1) total number of augmented data is 6 for one image
        # 6 types of augmentation means pca, horizontal and vertical flip, 3 rotation
        # if you specify range(2) total number of augmented data is 12 for one image (or multiply it as often as you want)
        for color in range(2):
            img_color = pca_color_augmentation(img)
            color_name = img_split[0] + '-color' + str(color)
            color_jpg = color_name + '.jpg'
            new_path = os.path.join(augmented_img_path, color_jpg)  # join with augmented image path
            lines = [color_jpg, ',', str(x1), ',', str(y1), ',', str(x2), ',', str(y2), ',', class_name, '\n']
            pwd_lines.append(lines)
            if not os.path.isfile(new_path):
                cv2.imwrite(new_path, img_color)

            # for horizontal and vertical flip
            f_points = [0, 1]
            for f in f_points:
                f_img = cv2.flip(img_color, f)
                h, w = img_color.shape[:2]

                if f == 1:
                    f_x1 = w - x2
                    f_y1 = y1
                    f_x2 = w - x1
                    f_y2 = y2
                    f_str = 'f1'
                elif f == 0:
                    f_x1 = x1
                    f_y1 = h - y2
                    f_x2 = x2
                    f_y2 = h - y1
                    f_str = 'f0'

                new_name = color_name + '-' + f_str + '.jpg'
                new_path = os.path.join(augmented_img_path, new_name)

                lines = [new_name, ',', str(f_x1), ',', str(f_y1), ',', str(f_x2), ',', str(f_y2), ',', class_name,
                         '\n']
                pwd_lines.append(lines)
                if not os.path.isfile(new_path):
                    cv2.imwrite(new_path, f_img)


            # for angle 180
            img_180 = cv2.flip(img_color, -1)
            ang_x1 = w - x2
            ang_y1 = h - y2
            ang_x2 = w - x1
            ang_y2 = h - y1
            new_name_180 = color_name + '-' + 'rotate_180' + '.jpg'
            new_path_180 = os.path.join(augmented_img_path, new_name_180)
            lines_180 = [new_name_180, ',', str(ang_x1), ',', str(ang_y1), ',', str(ang_x2), ',', str(ang_y2), ',',
                         class_name, '\n']
            pwd_lines.append(lines_180)
            if not os.path.isfile(new_path_180):
                cv2.imwrite(new_path_180, img_180)


# print(pwd_lines)
if len(pwd_lines) > 0:
    with open(textfile_out_path, 'w') as f:
        for line in pwd_lines:
            f.writelines(line)

print('End')