forked from tensorflow/models
-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathpredict_icdar15.py
185 lines (148 loc) · 6.42 KB
/
predict_icdar15.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
import numpy as np
import os
import sys
import time
import numpy as np
import tensorflow as tf
print tf.__file__ ###
from collections import defaultdict
from PIL import Image
sys.path.append("..")
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
from utils import label_map_util
PATH_TO_CKPT = '/world/data-c9/censhusheng/train_tf_text_loc/inference_graph-2.pb'
detection_graph = tf.Graph()
with detection_graph.as_default():
od_graph_def = tf.GraphDef()
with tf.gfile.GFile(PATH_TO_CKPT, 'rb') as fid:
serialized_graph = fid.read()
od_graph_def.ParseFromString(serialized_graph)
tf.import_graph_def(od_graph_def, name='')
PATH_TO_LABELS = 'data/text_label_map.pbtxt'
NUM_CLASSES = 1
label_map = label_map_util.load_labelmap(PATH_TO_LABELS)
categories = label_map_util.convert_label_map_to_categories(label_map, max_num_classes=NUM_CLASSES, use_display_name=True)
category_indexs = label_map_util.create_category_index(categories)
def resize(image, max_size, scale=1.0):
width, height = image.size
new_width = int(width * scale)
new_height = int(height * scale)
# meet the limit of max_size
if width > height and width > max_size:
new_width = max_size
new_height = int(height * new_width/ width)
if height > width and height > max_size:
new_height = max_size
new_width = int(width * new_height/height)
# round to multiple of 32
if new_width %32 != 0:
new_width = int(new_height/32) *32
if new_height %32 != 0:
new_height = int(new_height/32) *32
print 'resize:', (width, height), '->', (new_width, new_height) ###
return image.resize((new_width, new_height), Image.BILINEAR)
def load_image_into_numpy_array(input_image, max_size=1280, scale=1.0):
image = resize(input_image, max_size, scale)
(im_width, im_height) = image.size
return np.array(image).reshape(
(im_height, im_width, 3)).astype(np.uint8)
import math
def rbox_2_polygon(xc, yc, w, h, rad):
if w <1 or h <1:
return None
w2 = w/2
h2 = h/2
m_rot = np.array([[np.cos(rad), -np.sin(rad)],
[np.sin(rad), np.cos(rad)]], dtype=np.float32)
pts_ = np.array([[-w2,-h2],
[w2,-h2],
[w2,h2],
[-w2,h2]], dtype=np.float32)
pts = np.dot(m_rot, pts_.T).T
pts[:,0] = pts[:,0] + xc
pts[:,1] = pts[:,1] + yc
return pts.flatten().tolist()
sys.path.append(os.path.expanduser('~/github/EAST/'))
import lanms
print lanms
data_dir = '/world/data-c7/censhusheng/data/icdar2015-Incidental_Scene_Text/test_images/'
output_dir = '/world/data-c7/censhusheng/data/icdar2015-Incidental_Scene_Text/script_test/test'
config = tf.ConfigProto(
inter_op_parallelism_threads=2,
intra_op_parallelism_threads=2,
gpu_options = tf.GPUOptions(
per_process_gpu_memory_fraction = 0.5
)
)
with detection_graph.as_default():
with tf.Session(graph=detection_graph, config=config) as sess:
for i, fn in enumerate(os.listdir(data_dir)):
print '\n', i, os.path.join(data_dir, fn) ###
t_ = time.time()
image = Image.open(os.path.join(data_dir, fn))
width, height = image.size
polys = []
scales = [1.2]
for s in scales:
image_np = load_image_into_numpy_array(image, 2000, s)
new_width, new_height = image.size
ratio_x = float(width)/image_np.shape[1]
ratio_y = float(height)/image_np.shape[0]
print 'preprocss time:', time.time() - t_ ###
image_np_expanded = np.expand_dims(image_np, axis=0)
image_tensor = detection_graph.get_tensor_by_name('image_tensor:0')
boxes = detection_graph.get_tensor_by_name('detection_boxes:0')
scores = detection_graph.get_tensor_by_name('detection_scores:0')
classes = detection_graph.get_tensor_by_name('detection_classes:0')
num_detections = detection_graph.get_tensor_by_name('num_detections:0')
# Actual detection.
t0 = time.time()
(boxes, scores, classes, num_detections) = sess.run(
[boxes, scores, classes, num_detections],
feed_dict={image_tensor: image_np_expanded})
print 'net time:', time.time() - t0, 'num_detections:', num_detections ###
# convert rbox to polygon
t0 = time.time()
boxes = np.squeeze(boxes)
yc = (boxes[:,0] + boxes[:,2]) / 2
xc = (boxes[:,1] + boxes[:,3]) / 2
h = boxes[:,2] - boxes[:,0]
w = boxes[:,3] - boxes[:,1]
scores = np.squeeze(scores)
score_thresh = 0.5
for i in range(num_detections[0]):
if scores[i] < score_thresh:
continue
p = rbox_2_polygon(xc[i], yc[i], w[i], h[i], boxes[i,4])
if p is not None:
for i in range(8):
ratio = ratio_x if i%2 ==0 else ratio_y
p[i] *= ratio
polys.append(p + [scores[i]])
print 'convert time:', time.time() - t0 ###
polys = np.array(polys, dtype=np.float32)
# lanms
nms_thresh = 0.2
t0 = time.time()
polys = lanms.merge_quadrangle_n9(polys, nms_thresh)
nms_keep_thresh = 1.0
if polys.shape[0] > 0:
remain_index = np.where(polys[:,8] > nms_keep_thresh)[0]
polys = polys[remain_index]
print 'nms_time:', time.time() - t0, 'boxes:', polys.shape[0] ###
# save results
res_fn = 'res_' + os.path.splitext(fn)[0] + '.txt'
with open(os.path.join(output_dir, res_fn), 'w') as f_res:
for p in polys:
f_res.write("{:d},{:d},{:d},{:d},{:d},{:d},{:d},{:d}\r\n".format(
int(p[0]),
int(p[1]),
int(p[2]),
int(p[3]),
int(p[4]),
int(p[5]),
int(p[6]),
int(p[7]),
))
print os.path.join(output_dir, res_fn)
print 'total time:', time.time() - t_