-
Notifications
You must be signed in to change notification settings - Fork 0
/
landsat_test.py
240 lines (206 loc) · 8.55 KB
/
landsat_test.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
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
import os
import time
import argparse
import torch
from torchvision import transforms
import numpy as np
from PIL import Image
import sys
from model import HRcloudNet
import cv2
import re
from util.saliency_metric import cal_mae, cal_fm, cal_sm, cal_em, cal_wfm
def main():
parser = argparse.ArgumentParser(description='High_Resolution_cloud_net')
parser.add_argument('--d', default="38", type=str)
parser.add_argument('--m', default="best", type=str)
parser.add_argument('--gpu', default=0, type=int)
args = parser.parse_args()
DS = args.d
mdl = 'HRcloud_' + args.m
os.makedirs('testdata', exist_ok=True)
predict_multi(DS, mdl, args.gpu)
joint(DS, mdl)
test_metric_score(DS,mdl)
def subtest(DS, gpu):
mdl = "HRcloud_" + "best"
os.makedirs('testdata', exist_ok=True)
predict_multi(DS, mdl, gpu)
joint(DS, mdl)
test_metric_score(DS, mdl)
class test_dataset:
def __init__(self, image_root, gt_root):
self.img_list = [os.path.splitext(f)[0] for f in os.listdir(gt_root) if f.endswith('.png')]
self.image_root = image_root
self.gt_root = gt_root
self.transform = transforms.Compose([
transforms.ToTensor(),
])
self.gt_transform = transforms.ToTensor()
self.size = len(self.img_list)
self.index = 0
def load_data(self):
image = self.binary_loader(os.path.join(self.image_root,self.img_list[self.index]+ '.png'))
gt = self.binary_loader(os.path.join(self.gt_root,self.img_list[self.index] + '.png'))
self.index += 1
return image, gt
def rgb_loader(self, path):
with open(path, 'rb') as f:
img = Image.open(f)
return img.convert('RGB')
def binary_loader(self, path):
with open(path, 'rb') as f:
img = Image.open(f)
return img.convert('L')
def test_metric_score(DS, mdl):
p_dataset = DS + '_LandSat_8'
dataset_path = '/idata/cloud/' + DS + '_large_gt/'
if DS == 'CH':
dataset = DS + '_Test_img'
else:
dataset = DS + 'img'
dataset_path_pre = './testdata/' + dataset + mdl + '_large/'
sal_root = dataset_path_pre
gt_root = dataset_path
test_loader = test_dataset(sal_root, gt_root)
mae, fm, sm, em, wfm= cal_mae(), cal_fm(test_loader.size), cal_sm(), cal_em(), cal_wfm()
for i in range(test_loader.size):
sal, gt = test_loader.load_data()
if sal.size != gt.size:
x, y = gt.size
sal = sal.resize((x, y))
gt = np.asarray(gt, np.float32)
gt /= (gt.max() + 1e-20)
gt[gt > 0.5] = 1
gt[gt != 1] = 0
res = sal
res = np.array(res)
if res.max() == res.min():
res = res/255
else:
res = (res - res.min()) / (res.max() - res.min())
mae.update(res, gt)
sm.update(res,gt)
fm.update(res, gt)
em.update(res,gt)
wfm.update(res,gt)
MAE = mae.show()
maxf, meanf,_,_ = fm.show()
sm = sm.show()
em = em.show()
wfm = wfm.show()
print('dataset: {} MAE: {:.4f} maxF: {:.4f} avgF: {:.4f} wfm: {:.4f} Sm: {:.4f} Em: {:.4f}'.format(p_dataset, MAE, maxf,meanf,wfm,sm,em))
def joint(DS, mdl):
if DS == 'CH':
dataset = DS + '_Test_img'
else:
dataset = DS + 'img'
pic_path = './testdata/' + dataset + mdl +'/'
pic_target = './testdata/' + dataset + mdl +'_large/'
if not os.path.exists(pic_target):
os.mkdir(pic_target)
large_picture_names = os.listdir(pic_path)
if DS =='spars':
large_picture_names = [large_picture_names[15:] for large_picture_names in large_picture_names]
else:
large_picture_names = [large_picture_names[-44:] for large_picture_names in large_picture_names]
large_picture_names = set(large_picture_names)
# print(large_picture_names)
picture_names = os.listdir(pic_path)
# 随便一张图的的地址,用来查询小图长和宽
width, length, depth = 352, 352, 3
if len(picture_names)==0:
print("none")
else:
num = 0
for pic in large_picture_names:
# 获取文件夹中的所有文件名
filenames = os.listdir(pic_path)
# print(pic)
# 遍历文件名列表,并查找包含目标序列的文件名
if DS == 'spars':
num_width = 3
num_length= 3
else:
matching_filenames = [filename for filename in filenames if pic in filename]
a = 0
b = 0
for filename in matching_filenames:
w = int(filename.split("_")[2])
l = int(filename.split("_")[4])
if w > a:
a = w
if l > b:
b = l
num_width = a
num_length= b
num += 1
splicing_pic = np.zeros((num_width*width, num_length*length, depth))
for idx in range(0, 1):
k = 0
splicing_pic = np.zeros((num_width*width, num_length*length, depth))
for i in range(1, num_width+1):
for j in range(1, num_length+1):
k += 1
img_part = cv2.imread(pic_path + 'patch_{}_{}_by_{}_'.format(k, i, j)+pic,1)
splicing_pic[ width*(i-1) : width*i, length*(j-1) : length*j, :] = img_part
cv2.imwrite(pic_target + pic, splicing_pic)
print(num)
print("done!!!")
def time_synchronized():
torch.cuda.synchronize() if torch.cuda.is_available() else None
return time.time()
def predict_multi(DS, mdl, gpu):
if DS == 'CH':
dataset = DS + '_Test_img'
else:
dataset = DS + 'img'
test_data = '/data/cloud/' + dataset#
to_test = {'test':test_data}
classes = 1 # exclude background
weights_path = './result/gpu_' + str(gpu) +'/' + mdl + ".pth"
ckpt_path = './testdata/' + dataset + mdl
if not os.path.exists(ckpt_path):
os.mkdir(ckpt_path)
assert os.path.exists(weights_path), f"weights {weights_path} not found."
mean = (0.342, 0.413, 0.359) # Trian
std = (0.085, 0.094, 0.091)
# get devices
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
print("using {} device.".format(device))
model = HRcloudNet()
model.load_state_dict(torch.load(weights_path, map_location='cpu'))
model.to(device)
model.eval()
with torch.no_grad():
for name, root in to_test.items():
# 获取图片名称list
img_list = [os.path.splitext(f)[0] for f in os.listdir(root) if f.endswith('.png')]
# 开始计时
t_start = time_synchronized()
# 图像处理
data_transform = transforms.Compose([transforms.ToTensor(),
transforms.Normalize(mean=mean, std=std)])
for idx, img_name in enumerate(img_list):
# print ('predicting for %s: %d / %d' % (name, idx + 1, len(img_list)))
original_img = Image.open(os.path.join(root, img_name +'.png')).convert('RGB')
img = data_transform(original_img)
img = torch.unsqueeze(img, dim=0)
img_height, img_width = img.shape[-2:]
init_img = torch.zeros((1, 3, img_height, img_width), device=device)
model(init_img)
output = model(img.to(device))
prediction = output['out'].argmax(1).squeeze(0)
prediction = prediction.to("cpu").numpy().astype(np.uint8)
# 将前景对应的像素值改成255(白色)
prediction[prediction == 1] = 255
# 将不敢兴趣的区域像素设置成0(黑色)
# prediction[roi_img == 0] = 0
mask = Image.fromarray(prediction)
# mask.save(os.path.join(ckpt_path, exp_name, img_name + '.png'))
mask.save(os.path.join(ckpt_path, img_name + '.png'))
# 预测用时
t_end = time_synchronized()
# print("inference time: {}".format(t_end - t_start))
if __name__ == '__main__':
main()