ripenet/data_processor.py at main · personion222/ripenet · 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
from math import sin, cos, radians
from multiprocessing import Pool
from os.path import basename
from xml.dom import minidom
from os import listdir
import numpy as np
import effects
import random
import time
import qoi


def rot_point(poi, ori, angle):
    rads = radians(angle)
    rel_poi = (poi[0] - ori[0], poi[1] - ori[1])
    rotated = (rel_poi[0] * cos(rads) - rel_poi[1] * sin(rads),
               -(rel_poi[1] * cos(rads) + rel_poi[0] * sin(rads)))

    out = np.round((rotated[0] + ori[0], rotated[1] + ori[1]))

    return out


def scale_point(poi, ori, scale):
    rel_poi = (poi[0] - ori[0], poi[1] - ori[1])
    scaled = (rel_poi[0] * scale,
              rel_poi[1] * scale)

    out = np.round((scaled[0] + ori[0], scaled[1] + ori[1]))

    return out


def process_ann(ann):
    ann_name = ann.split('.')

    dom = minidom.parse(f"{ANN_DIR}/{ann}")
    file_path = dom.getElementsByTagName("path")[0].firstChild.nodeValue

    img_dims = dom.getElementsByTagName("size")[0]
    img_width = int(img_dims.getElementsByTagName("width")[0].firstChild.nodeValue)
    img_height = int(img_dims.getElementsByTagName("height")[0].firstChild.nodeValue)
    c = (img_width // 2, img_height // 2)

    bndbox = dom.getElementsByTagName("bndbox")[0]
    xmin = int(bndbox.getElementsByTagName("xmin")[0].firstChild.nodeValue)
    ymin = int(bndbox.getElementsByTagName("ymin")[0].firstChild.nodeValue)
    xmax = int(bndbox.getElementsByTagName("xmax")[0].firstChild.nodeValue)
    ymax = int(bndbox.getElementsByTagName("ymax")[0].firstChild.nodeValue)

    box_points = np.array(((xmin, ymin), (xmin, ymax), (xmax, ymin), (xmax, ymax)))

    img_name = basename(file_path).split('.')
    img_raw = qoi.read(file_path, 3)
    qoi.write(f"{ANN_IMG}/{basename(file_path)}", img_raw)
    dom.getElementsByTagName("path")[0].firstChild.nodeValue = f"{ANN_IMG}/{basename(file_path)}"

    for i in range(DUP_VAL):
        img_name_new = f"{img_name[0]}-{i}.{img_name[1]}"
        img_dir = f"{ANN_IMG}/{img_name_new}"
        if i > 0:
            angle = random.randint(0, 359)
            bgcol = np.full(3, np.random.randint(0, 128))
            scale = np.random.uniform(0.6, 1.0)
            shift = np.random.randint(-12, 12, size=2)

            mod_img = effects.rotate_img(img_raw, angle, bgcol)
            mod_img = effects.zoom_img(mod_img, scale, bgcol)
            mod_img = effects.shift_img(mod_img, shift, bgcol)

            noise_fac = random.randint(0, 16)
            mod_img = effects.add_noise(mod_img, noise_fac)

            box_points_rot = np.empty(box_points.shape, dtype=np.uint8)

            for idx, point in enumerate(box_points):
                box_points_rot[idx] = rot_point(point, c, angle)

            xmin_proc, ymin_proc = np.min(box_points_rot, axis=0)
            xmax_proc, ymax_proc = np.max(box_points_rot, axis=0)

            xmin_proc, ymin_proc = scale_point((xmin_proc, ymin_proc), c, scale)
            xmax_proc, ymax_proc = scale_point((xmax_proc, ymax_proc), c, scale)

            xmin_proc += shift[0]
            xmax_proc += shift[0]
            ymin_proc += shift[1]
            ymax_proc += shift[1]

            xmin_proc = round(np.clip(xmin_proc, 0, img_width))
            ymin_proc = round(np.clip(ymin_proc, 0, img_height))
            xmax_proc = round(np.clip(xmax_proc, 0, img_width))
            ymax_proc = round(np.clip(ymax_proc, 0, img_height))

            if xmin_proc >= xmax_proc or ymin_proc >= ymax_proc:
                continue

            bndbox.getElementsByTagName("xmin")[0].firstChild.nodeValue = str(xmin_proc)
            bndbox.getElementsByTagName("ymin")[0].firstChild.nodeValue = str(ymin_proc)
            bndbox.getElementsByTagName("xmax")[0].firstChild.nodeValue = str(xmax_proc)
            bndbox.getElementsByTagName("ymax")[0].firstChild.nodeValue = str(ymax_proc)

            qoi.write(img_dir, mod_img)

            dom.getElementsByTagName("filename")[0].firstChild.nodeValue = img_name_new
            dom.getElementsByTagName("path")[0].firstChild.nodeValue = img_dir

        dom.writexml(open(f"{ANN_SAVE}/{ann_name[0]}-{i}.{ann_name[1]}", 'w'))

    dom.unlink()


NOISE_MAX = 36
DUP_VAL = 25

ANN_DIR = "verified/ann"
ANN_SAVE = "fab_qoi/ann"
ANN_IMG = "fab_qoi/img"

annotations = listdir(ANN_DIR)

start = time.time()
with Pool() as p:
    p.map(process_ann, annotations)

print(f"finished in {round(time.time() - start, 2)} secs")