plugin-cloud-cover/module.py at main · waggle-sensor/plugin-cloud-cover · 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
127
128
import cv2, torch, torchvision, math, shapely
import numpy as np
import torch.nn.functional as F
from shapely.geometry import Polygon, Point
from skimage import measure


def circular_sector(center_x, center_y, radius, start_angle, end_angle, num_segments=36):
    angles = [start_angle + (float(i) / num_segments) * (end_angle - start_angle) for i in range(num_segments + 1)]

    points = []
    points.append((center_x, center_y))
    for angle in angles:
      rad = math.radians(angle)
      x = center_x + radius * math.cos(rad)
      y = center_y + radius * math.sin(rad)
      points.append((x, y))
    points.append((center_x, center_y))
    return Polygon(points)


def additional_information(copy_pred, pred):
    new_data = copy_pred.copy() * 255
    mask = np.asarray(np.where(new_data > 0)).T
    mask = [Point(x, y) for x, y in mask]

    level = 127
    contours = measure.find_contours(new_data, level)
    w, h = new_data.shape
    r = int((w/2) * (3/4))
    circle = Point(int(w/2), int(h/2)).buffer(r, resolution=10)
    circlepolygon = Polygon(circle)
    true_count = sum(circlepolygon.contains(point) for point in mask)
    ratio = true_count / (copy_pred.shape[0] * copy_pred.shape[1])
    fpoints = []

    conf = torch.max(pred[0][0], pred[0][1]).detach().cpu().numpy()

    if conf.mean() > 0 and conf.mean() < 5 and true_count > 500:

        points = circular_sector(int(w/2), int(h/2), r, 360, 0, num_segments=8)
        coords = np.array(points.exterior.coords[1:-1])

        c = 0
        for i in range(len(coords)-1):

            triangle = np.array([[int(w/2), int(h/2)], coords[i], coords[i+1], [int(w/2), int(h/2)]])
            tripolygon = Polygon(triangle)
            true_count = sum(tripolygon.contains(point) for point in mask)

            if true_count <= c:
                continue
            else:
                c = true_count

                tempimage = np.zeros((256, 256))
                if i < 2:
                    for k in range(128, 256):
                        for l in range(128):
                            if Point(k, l).within(tripolygon) and copy_pred[k][l] == True:
                                tempimage[k][l] = True

                elif i < 4:
                    for k in range(128):
                        for l in range(128):
                            if Point(k, l).within(tripolygon) and copy_pred[k][l] == True:
                                tempimage[k][l] = True

                elif i < 6:
                    for k in range(128):
                        for l in range(128, 256):
                            if shapely.within(Point(k, l), tripolygon):
                                tempimage[k][l] = True

                else:
                    for k in range(128, 256):
                        for l in range(128, 256):
                            if Point(k, l).within(tripolygon) and copy_pred[k][l] == True:
                                tempimage[k][l] = True

                fcontours = measure.find_contours(tempimage)
                if len(fcontours) == 0:
                    fpoint = shapely.centroid(tripolygon)
                    fpoints.append(fpoint.xy)
                for fcontour in fcontours:
                    fcoords = measure.approximate_polygon(fcontour, tolerance=2.5)
                    ffcoords = []
                    for cols in fcoords:
                        ffcoords.append((float(cols[1]), float(cols[0])))
                    fpoint = shapely.centroid(Polygon(ffcoords))
                    fpoints.append(fpoint.xy)

    return ratio, fpoints


class InferMain:
    def __init__(self):
        self.net = torch.load('new_cloudcover_unet.pth')
        self.device = 'cuda:0' if torch.cuda.is_available() else 'cpu'
        self.net.to(device=self.device).eval()
        self.transform = torchvision.transforms.Compose([
        #     torchvision.transforms.Resize((256, 256)),
            torchvision.transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
            ])

    def run(self, full_image):

        image = torch.from_numpy(cv2.resize(full_image, (256,256)).transpose(2, 0, 1)).type(torch.float32)
        image_transformed = self.transform(image).type(torch.uint8)
        # image = torchvision.io.decode_image(torchvision.io.encode_jpeg(image_transformed))


        with torch.no_grad():
            output = self.net(image.unsqueeze(0).to(self.device).float())

            copy_pred = output.argmax(axis=1).float()
            copy_pred = copy_pred.squeeze(0).detach().cpu().numpy()

            ratio, fpoints = additional_information(copy_pred, output)

            rgbimage = cv2.resize(full_image, (256, 256))
            result = cv2.cvtColor(copy_pred, cv2.COLOR_GRAY2BGR).astype(np.uint8)
            hi = np.concatenate((rgbimage, result), axis=1)


        return ratio, fpoints, hi