-
-
Notifications
You must be signed in to change notification settings - Fork 329
/
Copy pathonnx_transform.py
159 lines (131 loc) · 6.26 KB
/
onnx_transform.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
import sys
import onnx
import os
import argparse
import numpy as np
import cv2
import onnxruntime
import torch
from utils.tools import *
# from tool.utils import *
from model.build_model import Build_Model
from eval.evaluator import *
import config.yolov4_config as cfg
def convert_predbox(pred_bbox, test_input_size, org_img_shape, valid_scale):
"""
预测框进行过滤，去除尺度不合理的框
"""
pred_coor = xywh2xyxy(pred_bbox[:, :4])
pred_conf = pred_bbox[:, 4]
pred_prob = pred_bbox[:, 5:]
# (1)
# (xmin_org, xmax_org) = ((xmin, xmax) - dw) / resize_ratio
# (ymin_org, ymax_org) = ((ymin, ymax) - dh) / resize_ratio
# 需要注意的是，无论我们在训练的时候使用什么数据增强方式，都不影响此处的转换方式
# 假设我们对输入测试图片使用了转换方式A，那么此处对bbox的转换方式就是方式A的逆向过程
org_h, org_w = org_img_shape
resize_ratio = min(1.0 * test_input_size / org_w, 1.0 * test_input_size / org_h)
dw = (test_input_size - resize_ratio * org_w) / 2
dh = (test_input_size - resize_ratio * org_h) / 2
pred_coor[:, 0::2] = 1.0 * (pred_coor[:, 0::2] - dw) / resize_ratio
pred_coor[:, 1::2] = 1.0 * (pred_coor[:, 1::2] - dh) / resize_ratio
# (2)将预测的bbox中超出原图的部分裁掉
pred_coor = np.concatenate([np.maximum(pred_coor[:, :2], [0, 0]),
np.minimum(pred_coor[:, 2:], [org_w - 1, org_h - 1])], axis=-1)
# (3)将无效bbox的coor置为0
invalid_mask = np.logical_or((pred_coor[:, 0] > pred_coor[:, 2]), (pred_coor[:, 1] > pred_coor[:, 3]))
pred_coor[invalid_mask] = 0
# (4)去掉不在有效范围内的bbox
bboxes_scale = np.sqrt(np.multiply.reduce(pred_coor[:, 2:4] - pred_coor[:, 0:2], axis=-1))
scale_mask = np.logical_and((valid_scale[0] < bboxes_scale), (bboxes_scale < valid_scale[1]))
# (5)将score低于score_threshold的bbox去掉
classes = np.argmax(pred_prob, axis=-1)
scores = pred_conf * pred_prob[np.arange(len(pred_coor)), classes]
score_mask = scores > cfg.VAL["CONF_THRESH"]
mask = np.logical_and(scale_mask, score_mask)
coors = pred_coor[mask]
scores = scores[mask]
classes = classes[mask]
bboxes = np.concatenate([coors, scores[:, np.newaxis], classes[:, np.newaxis]], axis=-1)
return bboxes
def detect(session, image_src):
IN_IMAGE_H = session.get_inputs()[0].shape[2]
IN_IMAGE_W = session.get_inputs()[0].shape[3]
# Input
resized = cv2.resize(image_src, (IN_IMAGE_W, IN_IMAGE_H), interpolation=cv2.INTER_LINEAR)
img_in = cv2.cvtColor(resized, cv2.COLOR_BGR2RGB)
img_in = np.transpose(img_in, (2, 0, 1)).astype(np.float32)
img_in = np.expand_dims(img_in, axis=0)
img_in /= 255.0
print("Shape of the network input: ", img_in.shape)
# Compute
input_name = session.get_inputs()[0].name
outputs = session.run(None, {input_name: img_in})
bboxes = convert_predbox(outputs[-1], IN_IMAGE_H, (IN_IMAGE_H, IN_IMAGE_W), (0, np.inf))
bboxes_prd = nms(bboxes, cfg.VAL["CONF_THRESH"], cfg.VAL["NMS_THRESH"])
if bboxes_prd.shape[0] != 0:
boxes = bboxes_prd[..., :4]
class_inds = bboxes_prd[..., 5].astype(np.int32)
scores = bboxes_prd[..., 4]
visualize_boxes(image=image_src, boxes=boxes, labels=class_inds, probs=scores, class_labels=cfg.VOC_DATA["CLASSES"])
path = os.path.join(cfg.PROJECT_PATH, "save.jpg")
cv2.imwrite(path, image_src)
print("saved images : {}".format(path))
def transform_to_onnx(weight_file, batch_size, IN_IMAGE_H, IN_IMAGE_W):
model = Build_Model()
pretrained_dict = torch.load(weight_file, map_location=torch.device('cpu'))
model.load_state_dict(pretrained_dict)
input_names = ["input"]
output_names = ['boxes', 'confs']
dynamic = False
if batch_size <= 0:
dynamic = True
if dynamic:
x = torch.randn((1, 3, IN_IMAGE_H, IN_IMAGE_W), requires_grad=True)
onnx_file_name = "yolov4_-1_3_{}_{}_dynamic.onnx".format(IN_IMAGE_H, IN_IMAGE_W)
dynamic_axes = {"input": {0: "batch_size"}, "boxes": {0: "batch_size"}, "confs": {0: "batch_size"}}
# Export the model
print('Export the onnx model ...')
torch.onnx.export(model,
x,
onnx_file_name,
export_params=True,
opset_version=11,
do_constant_folding=True,
input_names=input_names, output_names=output_names,
dynamic_axes=dynamic_axes)
print('Onnx model exporting done')
return onnx_file_name
else:
x = torch.randn((batch_size, 3, IN_IMAGE_H, IN_IMAGE_W), requires_grad=True)
onnx_file_name = "yolov4_{}_3_{}_{}_static.onnx".format(batch_size, IN_IMAGE_H, IN_IMAGE_W)
# Export the model
print('Export the onnx model ...')
torch.onnx.export(model,
x,
onnx_file_name,
export_params=True,
do_constant_folding=True,
input_names=input_names, output_names=output_names,
)
print('Onnx model exporting done')
return onnx_file_name
def main(weight_file=None, image_path= None, batch_size= 1, IN_IMAGE_H= 416, IN_IMAGE_W= 416):
if batch_size <= 0:
onnx_path_demo = transform_to_onnx(weight_file, batch_size, IN_IMAGE_H, IN_IMAGE_W)
else:
# Transform to onnx as specified batch size
transform_to_onnx(weight_file, batch_size, IN_IMAGE_H, IN_IMAGE_W)
# Transform to onnx for demo
onnx_path_demo = transform_to_onnx(weight_file, 1, IN_IMAGE_H, IN_IMAGE_W)
session = onnxruntime.InferenceSession(onnx_path_demo)
print("The model expects input shape: ", session.get_inputs()[0].shape)
image_src = cv2.imread(image_path)
detect(session, image_src)
if __name__ == '__main__':
import os.path as osp
print("Converting to onnx and running demo ...")
PROJECT_PATH = osp.abspath(osp.dirname(__file__))
weight_file = osp.join(PROJECT_PATH, 'weight/best.pt')
image_path = osp.join(PROJECT_PATH, '000001.jpg')
main(weight_file=weight_file,image_path=image_path)