-
Notifications
You must be signed in to change notification settings - Fork 10
/
Copy pathinfer.py
270 lines (241 loc) · 11.5 KB
/
infer.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
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
#!/usr/bin/python3
# -*- coding: utf-8 -*-
# @Author : BADBADBADBOY
# @File : infer.py
# @Time : 2024/8/20 21:22
import time
import os
import cv2
import torch
import shutil
import math
import numpy as np
from utils.utils import get_affine_transform,bbox_decode,decode_by_ind,nms,bbox_post_process,get_package_installation_path,get_train_file
from openvino.runtime import Core, AsyncInferQueue
from utils.utils import order_points_new,draw_show_img,merge_images_horizontal
from config import *
class CardRotate(object):
K = K
input_h, input_w = TEST_H, TEST_W
def load_torch(self,model_path):
super(CardRotate,self).__init__()
if BASE_MODEL == "resnet":
from model.models_resnet import CardDetectionCorrectionModel
model = CardDetectionCorrectionModel(num_layers=NUM_LAYER, need_ftype=NEED_FTYPE)
elif BASE_MODEL == "lcnet":
from model.models_lcnet import CardDetectionCorrectionModel
model = CardDetectionCorrectionModel(ratio=MODEL_RATIO, need_ftype=NEED_FTYPE)
elif BASE_MODEL == "replcnet":
from model.models_replcnet import CardDetectionCorrectionModel
from model.rep_blocks import repvgg_model_convert
model = CardDetectionCorrectionModel(ratio=MODEL_RATIO, need_ftype=NEED_FTYPE)
if TEST_LOAD_TYPE == 'modelscope' and BASE_MODEL=='resnet':
model_dict = torch.load(model_path, map_location='cpu')['state_dict']
else:
model_dict = torch.load(model_path, map_location='cpu')
try:
model.load_state_dict(model_dict)
except:
new_model_dict = {}
for key in model.state_dict().keys():
new_model_dict[key] = model_dict['module.' + key]
model.load_state_dict(new_model_dict)
if BASE_MODEL == "replcnet":
model = repvgg_model_convert(model)
print("Convert rep model ok !!!")
self.infer_model = model
self.infer_model.eval()
self.infer_type = 'torch'
if torch.cuda.is_available() and USE_GPU:
self.infer_model = self.infer_model.cuda()
def load_openvino(self,openvino_model_path):
ie = Core()
model_ir = ie.read_model(model=openvino_model_path)
try:
compiled_model_ir = ie.compile_model(model=model_ir, device_name="CPU",
config={"PERFORMANCE_HINT": "LATENCY",
"CPU_RUNTIME_CACHE_CAPACITY": "0",
'CPU_THREADS_NUM': '4'})
except:
compiled_model_ir = ie.compile_model(model=model_ir, device_name="CPU",
config={"PERFORMANCE_HINT": "LATENCY",
"CPU_RUNTIME_CACHE_CAPACITY": "0"
})
self.infer_request = compiled_model_ir.create_infer_request()
self.infer_type = 'openvino'
def rotate(self,img):
pre_input = self.preprocess(img)
if self.infer_type == 'torch':
pre_out = self.forward(pre_input)
elif self.infer_type == 'openvino':
pre_out = self.forward_openvino(pre_input)
for key in pre_out['results'][0].keys():
pre_out['results'][0][key] = torch.from_numpy(pre_out['results'][0][key])
out = self.postprocess(pre_out)
return out
def onnx_export(self,img,onnx_path):
torch.onnx.export(self.infer_model, # 导出的模型
self.preprocess(img)['img'], # 输入数据
onnx_path, # 保存的文件名
export_params=True, # 是否导出模型参数
opset_version=11, # ONNX 版本号
do_constant_folding=True, # 是否执行常量折叠优化
input_names=['input'], # 输入名
output_names=['output']) # 输出名
def onnx2openvino(self,use_fp16 = True):
os.system(r"python {} "
"--input_model {} "
"--input_shape [-1,3,-1,-1] "
"--compress_to_fp16 {} "
"--output_dir {}".format(os.path.join(get_package_installation_path("openvino"),'openvino/tools/mo/mo.py'),'./model.onnx',use_fp16,'./openvino'))
def preprocess(self, img):
self.image = img
mean = np.array([0.408, 0.447, 0.470],dtype=np.float32).reshape(1, 1, 3)
std = np.array([0.289, 0.274, 0.278],dtype=np.float32).reshape(1, 1, 3)
height, width = img.shape[0:2]
inp_height, inp_width = self.input_h, self.input_w
c = np.array([width / 2., height / 2.], dtype=np.float32)
s = max(height, width) * 1.0
trans_input = get_affine_transform(c, s, 0, [inp_width, inp_height])
resized_image = cv2.resize(img, (width, height))
inp_image = cv2.warpAffine(
resized_image,
trans_input, (inp_width, inp_height),
flags=cv2.INTER_LINEAR)
inp_image = ((inp_image / 255. - mean) / std).astype(np.float32)
images = inp_image.transpose(2, 0, 1).reshape(1, 3, inp_height,inp_width)
if self.infer_type == 'torch':
images = torch.from_numpy(images)
if torch.cuda.is_available() and USE_GPU:
images = images.cuda()
meta = {
'c': c,
's': s,
'input_height': inp_height,
'input_width': inp_width,
'out_height': inp_height // 4,
'out_width': inp_width // 4
}
result = {'img': images, 'meta': meta}
return result
def distance(self, x1, y1, x2, y2):
return math.sqrt(pow(x1 - x2, 2) + pow(y1 - y2, 2))
def crop_image(self, img, position):
x0, y0 = position[0][0], position[0][1]
x1, y1 = position[1][0], position[1][1]
x2, y2 = position[2][0], position[2][1]
x3, y3 = position[3][0], position[3][1]
img_width = self.distance((x0 + x3) / 2, (y0 + y3) / 2, (x1 + x2) / 2,
(y1 + y2) / 2)
img_height = self.distance((x0 + x1) / 2, (y0 + y1) / 2, (x2 + x3) / 2,
(y2 + y3) / 2)
corners_trans = np.zeros((4, 2), np.float32)
corners_trans[0] = [0, 0]
corners_trans[1] = [img_width, 0]
corners_trans[2] = [img_width, img_height]
corners_trans[3] = [0, img_height]
transform = cv2.getPerspectiveTransform(position, corners_trans)
dst = cv2.warpPerspective(img, transform,
(int(img_width), int(img_height)))
return dst
def forward(self, input) :
pred = self.infer_model(input['img'])
return {'results': pred, 'meta': input['meta']}
def forward_openvino(self,input):
self.infer_request.infer([input['img']])
openvino_output = {}
if NEED_FTYPE:
openvino_output['hm'] = self.infer_request.get_output_tensor(0).data
openvino_output['cls'] = self.infer_request.get_output_tensor(1).data
openvino_output['ftype'] = self.infer_request.get_output_tensor(2).data
openvino_output['wh'] = self.infer_request.get_output_tensor(3).data
openvino_output['reg'] = self.infer_request.get_output_tensor(4).data
else:
openvino_output['hm'] = self.infer_request.get_output_tensor(0).data
openvino_output['cls'] = self.infer_request.get_output_tensor(1).data
openvino_output['wh'] = self.infer_request.get_output_tensor(2).data
openvino_output['reg'] = self.infer_request.get_output_tensor(3).data
return {'results': [openvino_output], 'meta': input['meta']}
def postprocess(self, inputs):
output = inputs['results'][0]
meta = inputs['meta']
wh = output['wh']
reg = output['reg']
hm = output['hm'].sigmoid_()
angle_cls = output['cls'].sigmoid_()
if NEED_FTYPE:
ftype_cls = output['ftype'].sigmoid_()
bbox, inds = bbox_decode(hm, wh, reg=reg, K=self.K)
angle_cls = decode_by_ind(angle_cls, inds, K=self.K).detach().cpu().numpy()
if NEED_FTYPE:
ftype_cls = decode_by_ind(ftype_cls, inds,K=self.K).detach().cpu().numpy().astype(np.float32)
bbox = bbox.detach().cpu().numpy()
for i in range(bbox.shape[1]):
bbox[0][i][9] = angle_cls[0][i]
if NEED_FTYPE:
bbox = np.concatenate((bbox, np.expand_dims(ftype_cls, axis=-1)),axis=-1)
bbox = nms(bbox, 0.3)
bbox = bbox_post_process(bbox.copy(), [meta['c']],[meta['s']], meta['out_height'],meta['out_width'])
res = []
angle = []
sub_imgs = []
ftype = []
score = []
center = []
corner_left_right = []
for idx, box in enumerate(bbox[0]):
if box[8] > OBJ_SCORE:
angle.append(int(box[9]))
res.append(box[0:8])
box8point = np.array(box[0:8]).reshape(4,2).astype(np.int32)
corner_left_right.append([box8point[:,0].min(),box8point[:,1].min(),box8point[:,0].max(),box8point[:,1].max()])
sub_img = self.crop_image(self.image,res[-1].copy().reshape(4, 2))
if angle[-1] == 1:
sub_img = cv2.rotate(sub_img, 2)
if angle[-1] == 2:
sub_img = cv2.rotate(sub_img, 1)
if angle[-1] == 3:
sub_img = cv2.rotate(sub_img, 0)
sub_imgs.append(sub_img)
if NEED_FTYPE:
ftype.append(int(box[12]))
score.append(box[8])
center.append([box[10],box[11]])
result = {
"POLYGONS": np.array(res),
"BBOX": np.array(corner_left_right),
"SCORES": np.array(score),
"OUTPUT_IMGS": sub_imgs,
"LABELS": np.array(angle),
"LAYOUT": np.array(ftype),
"CENTER": np.array(center)
}
return result
if __name__ == "__main__":
# get_train_file(det_dir = "/src/notebooks/trainimg/image",pre_img_dir='./pre_img',pre_gt_dir='./pre_gt',model_path=TEST_MODEL_PATH)
img = cv2.imread(r'../SyntheticCards_train100k/data/place_798_aug_0.jpg')
# img = cv2.imread(r'./pre_img/yyzz_img_16_rotated_0deg.jpg')
rotate_bin = CardRotate()
if TEST_TYPE == "test_torch":
rotate_bin.load_torch(TEST_MODEL_PATH)
elif TEST_TYPE == "test_openvino":
rotate_bin.load_openvino(TEST_MODEL_PATH)
elif TEST_TYPE == "torch2openvino":
rotate_bin.load_torch(TEST_MODEL_PATH)
print("torch模型转换onnx模型中......")
rotate_bin.onnx_export(img,"./model.onnx")
print("onnx模型转换完成 ！！！")
print("onnx模型转换openvino模型中......")
rotate_bin.onnx2openvino(use_fp16=USE_FP16)
print("openvino模型转换成功 ！！！")
os.remove("./model.onnx")
if TEST_TYPE == "test_torch" or TEST_TYPE == "test_openvino":
t_sum = 0
for i in range(1):
t = time.time()
out = rotate_bin.rotate(img)
t_sum += (time.time() - t)
print("time avg:{}".format(t_sum/1))
draw_show_img(img.copy(), out)
merge_images_horizontal([img] + out['OUTPUT_IMGS'],"./pp4_rotate_show.jpg")
cv2.imwrite(r'./rotate_img.jpg',out['OUTPUT_IMGS'][0])