BuildingChangeDetector/process_mark.py at master · LiuDongjing/BuildingChangeDetector · 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
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
# -*- coding: utf-8 -*-
"""
    根据手工标注的标签生成npy文件，每个npy文件保存了一个二维数组(width, height, 8+1)，
    前8个通道是图像数据，最后一个通道是标签
"""
import os
import sys
import glob
import json

import tqdm
import skimage.io
import numpy as np
import matplotlib.pyplot as plt

def find_bnd(img):
    """
    从标签图像中找到标注的区域(矩形)，标签图像是与卫星图大小一致的RGBA图像，未标注的
    区域为黑色，标注的区域如果是房子变化则为红色，否则为透明(四通道数值均为0)。
    """
    r_s = -1
    r_e = -1
    c_s = -1
    c_e = -1
    fg = False
    r_m = -1
    c_m = -1
    for i in range(0, img.shape[0], 256):
        if fg:
            break
        for j in range(0, img.shape[1], 256):
            if img[i, j, 0] > 0 or img[i,j,3] == 0:
                r_m = i
                c_m = j
                fg = True
                break
    if r_m < 0 or c_m < 0:
        return r_s, r_e, c_s, c_e
    r_s = r_m
    c_s = c_m
    while img[r_s, c_m, 0] > 0 or img[r_s, c_m, 3] == 0:
        r_s -= 1
    r_s += 1
    while img[r_m, c_s, 0] > 0 or img[r_m, c_s, 3] == 0:
        c_s -= 1
    c_s += 1
    r_e = r_m+1
    c_e = c_m+1
    while img[r_e, c_m, 0] > 0 or img[r_e, c_m, 3] == 0:
        r_e += 1
    while img[r_m, c_e, 0] > 0 or img[r_m, c_e, 3] == 0:
        c_e += 1
    return r_s, r_e, c_s, c_e

def prepare_end2end_data(path15, path17, input_dir, output_dir, base=0, suffix='p2'):
    """
    将标注的标签图像转换成npy文件。
    """
    if not os.path.exists(output_dir):
        os.makedirs(output_dir)
    im15 = skimage.io.imread(path15).astype(np.float32)
    im17 = skimage.io.imread(path17).astype(np.float32)
    masks = glob.glob(os.path.join(input_dir, '*.tif'))
    data_dir = output_dir
    if not os.path.exists(data_dir):
        os.makedirs(data_dir)
    for mp in tqdm.tqdm(masks):
        msk = skimage.io.imread(mp)
        r_s, r_e, c_s, c_e = find_bnd(msk)
        if r_s < 0:
            print('%s无效!'%mp, file=sys.stderr)
            continue
        d15 = im15[r_s:r_e, c_s:c_e, :]
        d17 = im17[r_s:r_e, c_s:c_e, :]
        m = msk[r_s:r_e, c_s:c_e, 0]
        lab = m > 0
        lab = lab.astype(d15.dtype)
        lab = np.expand_dims(lab, 2)
        d = np.concatenate([d15, d17, lab], 2)
        vp = d.shape[0]//2
        hp = d.shape[1]//2
        lst = [d[:vp, :hp, :], d[vp:, :hp, :], d[:vp, hp:, :], d[vp:, hp:, :]]
        cord = [(r_s, c_s), (r_s+vp, c_s), (r_s, c_s+hp), (r_s+vp, c_s+hp)]
        _, n = os.path.split(mp)
        n, _ = os.path.splitext(n)
        for k in range(len(lst)):
            r, c = cord[k]
            dp = os.path.join(data_dir, '%d_%d_%d#%d_%s.npy'%(base, k, r, c, suffix))
            np.save(dp, lst[k])
        base += 1

def end2end_split(data_dir, splits=None):
    """
    将npy文件划分成训练集、验证集和测试集
    """
    if splits is None:
        splits = [0.7, 0.9]
    assert splits[0] < splits[1]
    dat_all = glob.glob(os.path.join(data_dir, '*.npy'))
    np.random.shuffle(dat_all)
    for k in range(len(dat_all)):
        _, dat_all[k] = os.path.split(dat_all[k])
    mp = {}
    sp1 = int(len(dat_all)*splits[0])
    sp2 = int(len(dat_all)*splits[1])
    mp['train'] = dat_all[:sp1]
    mp['validation'] = dat_all[sp1:sp2]
    mp['test'] = dat_all[sp2:]
    if data_dir[-1] == '/' or data_dir[-1] == '\\':
        _, dir_name = os.path.split(data_dir[:-1])
    else:
        _, dir_name = os.path.split(data_dir)
    with open(os.path.join(data_dir, '%s_train_val_test.json'%dir_name), 'w') as file:
        file.write(json.dumps(mp))

def end2end_data_view(input_dir, part='validation'):
    """
    查看训练集、验证集或测试集里的图像
    """
    mp = glob.glob(os.path.join(input_dir, '*.json'))[0]
    mp = json.load(open(mp))
    paths = mp[part]
    for p in paths:
        t = os.path.join(input_dir, p)
        x = np.load(t)
        im15 = skimage.img_as_ubyte(x[:,:,:3].astype(np.uint16))
        im17 = skimage.img_as_ubyte(x[:,:,4:7].astype(np.uint16))
        msk = x[:,:,-1].astype(np.uint8)
        msk *= 90
        msk += 255-90
        msk = np.expand_dims(msk, 2)
        im15 = np.concatenate([im15, msk],2)
        im17 = np.concatenate([im17, msk],2)
        plt.subplot(1,2,1)
        plt.imshow(im15)
        plt.title('2015')
        plt.suptitle(p)
        plt.subplot(1,2,2)
        plt.imshow(im17)
        plt.title('2017')
        plt.show()

if __name__ == '__main__':
    prepare_end2end_data(
        '../../input/origin/2015p2-denoise-rgbn.tif',
        '../../input/origin/2017p2-denoise-rgbn.tif',
        '../../input/mark/p2_end2end_1102/',
        '../../input/mark/p2_test/', base=0, suffix='p2')
    end2end_split('../../input/mark/p2_test/')
    end2end_data_view('../../input/mark/p2_test/', part='validation')