-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathpredict.py
158 lines (138 loc) · 5.75 KB
/
predict.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
import argparse
import glob
import os.path
from copy import copy
import numpy
import paddle
import paddle.nn as nn
import cv2
import time
from PIL import Image
from model import *
import os
import warnings
from transforms import horizontal_flip, vertical_flip, resize
from utils import load_pretrained_model
warnings.filterwarnings('ignore')
def parse_args():
parser = argparse.ArgumentParser(description='Model testing')
parser.add_argument(
'--dataset_root',
dest='dataset_root',
help='The path of dataset root',
type=str,
default='data/data121607/data/test')
parser.add_argument(
'--pretrained',
dest='pretrained',
help='The pretrained of model',
type=str,
default='train_result/model/epoch_100/model.pdparams')
return parser.parse_args()
def predict(img, model, input_size=[256, 256]):
h_num, w_num = img.shape[0] // input_size[0], img.shape[1] // input_size[1]
blur_img = np.zeros(shape=img.shape, dtype=np.float32)
for h in range(h_num):
for w in range(w_num):
img_part = paddle.to_tensor(
img[h * input_size[0]:(h + 1) * input_size[0], w * input_size[1]:(w + 1) * input_size[1], :],
dtype=paddle.float32)
img_part = img_part.unsqueeze(0).transpose((0, 3, 1, 2))
img_out = model(img_part)
img_out = nn.functional.interpolate(img_out, size=input_size, mode="bilinear")
img_out = img_out * 255.0
img_out = paddle.clip(img_out, 0, 255)
img_out = img_out.squeeze()
img_out = paddle.transpose(img_out, [1, 2, 0])
img_out = img_out.numpy()
blur_img[h * input_size[0]:(h + 1) * input_size[0],
w * input_size[1]:(w + 1) * input_size[1],
:] = img_out
# 边缘
h_remain = img.shape[0] - h_num * input_size[0]
w_remain = img.shape[1] - w_num * input_size[1]
if h_remain != 0:
# 剩余高度
for w in range(w_num):
img_part = paddle.to_tensor(
img[-input_size[0]:, w * input_size[1]:(w + 1) * input_size[1], :],
dtype=paddle.float32
)
img_part = img_part.unsqueeze(0).transpose((0, 3, 1, 2))
img_out = model(img_part)
img_out = nn.functional.interpolate(img_out, size=input_size, mode="bilinear")
img_out = img_out * 255.0
img_out = paddle.clip(img_out, 0, 255)
img_out = img_out.squeeze()
img_out = paddle.transpose(img_out, [1, 2, 0])
img_out = img_out.numpy()
blur_img[-h_remain:, w * input_size[1]:(w + 1) * input_size[1], :] = img_out[-h_remain:, :, :]
if w_remain != 0:
# 剩余宽度
for h in range(h_num):
img_part = paddle.to_tensor(
img[h * input_size[0]:(h + 1) * input_size[0], -input_size[1]:, :],
dtype=paddle.float32
)
img_part = img_part.unsqueeze(0).transpose((0, 3, 1, 2))
img_out = model(img_part)
img_out = nn.functional.interpolate(img_out, size=input_size, mode="bilinear")
img_out = img_out * 255.0
img_out = paddle.clip(img_out, 0, 255)
img_out = img_out.squeeze()
img_out = paddle.transpose(img_out, [1, 2, 0])
img_out = img_out.numpy()
blur_img[h * input_size[0]:(h + 1) * input_size[0], -w_remain:, :] = img_out[:, -w_remain:, :]
if w_remain != 0 and h_remain != 0:
img_part = paddle.to_tensor(
img[-input_size[0]:, -input_size[1]:, :],
dtype=paddle.float32
)
img_part = img_part.unsqueeze(0).transpose((0, 3, 1, 2))
img_out = model(img_part)
img_out = nn.functional.interpolate(img_out, size=input_size, mode="bilinear")
img_out = img_out * 255.0
img_out = paddle.clip(img_out, 0, 255)
img_out = img_out.squeeze()
img_out = paddle.transpose(img_out, [1, 2, 0])
img_out = img_out.numpy()
blur_img[-h_remain:, -w_remain:, :] = img_out[-h_remain:, -w_remain:, :]
return blur_img
def main(args):
multi_output = False
model = MBCNN(64, multi_output) # MBCNN-light: model = MBCNN(32,multi_output)
if args.pretrained is not None:
load_pretrained_model(model, args.pretrained)
im_files = glob.glob(os.path.join(args.dataset_root, "images/*.jpg"))
input_size = [256, 256]
for i, im in enumerate(im_files):
start = time.time()
img = cv2.imread(im)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
ori_h,ori_w=img.shape[0],img.shape[1]
if ori_h > 1500 or ori_w > 1500:
img = resize(img, (ori_h *2//3, ori_w *2//3))
model.eval()
img = paddle.to_tensor(img)
img /= 255.0
blur_img = np.zeros(shape=img.shape, dtype=np.float32)
blur_img += 0.25 * predict(img, model, input_size)
h_img = horizontal_flip(img)
blur_img += 0.25 * predict(h_img, model, input_size)
vh_img = vertical_flip(h_img)
blur_img += 0.25 * horizontal_flip(predict(vh_img, model, input_size))
v_img = horizontal_flip(vh_img)
blur_img += 0.25 * horizontal_flip(predict(v_img, model, input_size))
blur_img=resize(blur_img,(ori_h,ori_w))
blur_img = np.clip(blur_img, 0, 255)
save_path = "output/"
if not os.path.exists(save_path):
os.makedirs(save_path)
blur_img = cv2.cvtColor(blur_img, cv2.COLOR_RGB2BGR)
cv2.imwrite(os.path.join(save_path, im.split('/')[-1]), blur_img)
end = time.time()
time_one = end - start
print('The running time of an image is : {:2f} s'.format(time_one))
if __name__ == '__main__':
args = parse_args()
main(args)