forked from shouxieai/tensorRT_Pro
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathapp_lesson.cpp
More file actions
198 lines (150 loc) · 5.87 KB
/
Copy pathapp_lesson.cpp
File metadata and controls
198 lines (150 loc) · 5.87 KB
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
#include <common/ilogger.hpp>
#include <infer/trt_infer.hpp>
#include <builder/trt_builder.hpp>
#include "app_yolo/yolo.hpp"
using namespace std;
static void lesson1(){
/** 模型编译，onnx到trtmodel **/
TRT::compile(
TRT::Mode::FP32, /** 模式, fp32 fp16 int8 **/
1, /** 最大batch size **/
"lesson1.onnx", /** onnx文件，输入 **/
"lesson1.fp32.trtmodel" /** trt模型文件，输出 **/
);
/** 加载编译好的引擎 **/
auto infer = TRT::load_infer("lesson1.fp32.trtmodel");
/** 设置输入的值 **/
infer->input(0)->set_to(1.0f);
/** 引擎进行推理 **/
infer->forward();
/** 取出引擎的输出并打印 **/
auto out = infer->output(0);
INFO("out.shape = %s", out->shape_string());
for(int i = 0; i < out->channel(); ++i)
INFO("%f", out->at<float>(0, i));
}
/** 动态batch **/
static void lesson2(){
int max_batch_size = 5;
/** 模型编译，onnx到trtmodel **/
TRT::compile(
TRT::Mode::FP32, /** 模式, fp32 fp16 int8 **/
max_batch_size, /** 最大batch size **/
"lesson1.onnx", /** onnx文件，输入 **/
"lesson1.fp32.trtmodel" /** trt模型文件，输出 **/
);
/** 加载编译好的引擎 **/
auto infer = TRT::load_infer("lesson1.fp32.trtmodel");
/** 设置输入的值 **/
/** 修改input的0维度为1，最大可以是5 **/
infer->input(0)->resize_single_dim(0, 2);
infer->input(0)->set_to(1.0f);
/** 引擎进行推理 **/
infer->forward();
/** 取出引擎的输出并打印 **/
auto out = infer->output(0);
INFO("out.shape = %s", out->shape_string());
}
/** 动态宽高-相对的，仅仅调整onnx输入大小为目的 **/
static void lesson3(){
TRT::set_layer_hook_reshape([](const string& name, const vector<int64_t>& shape)->vector<int64_t>{
INFO("name: %s, shape: %s", name.c_str(), iLogger::join_dims(shape).c_str());
return {-1, 25};
});
/** 模型编译，onnx到trtmodel **/
TRT::compile(
TRT::Mode::FP32, /** 模式, fp32 fp16 int8 **/
1, /** 最大batch size **/
"lesson1.onnx", /** onnx文件，输入 **/
"lesson1.fp32.trtmodel", /** trt模型文件，输出 **/
{{1, 1, 5, 5}} /** 对输入的重定义 **/
);
auto infer = TRT::load_infer("lesson1.fp32.trtmodel");
auto out = infer->output(0);
INFO("out.shape = %s", out->shape_string());
}
void lesson_cache1frame(){
iLogger::set_log_level(iLogger::LogLevel::Info);
auto model_file = "yolox_s.FP32.trtmodel";
auto onnx_file = "yolox_s.onnx";
if(not iLogger::exists(model_file)){
TRT::compile(
TRT::Mode::FP32, // FP32、FP16、INT8
16, // max batch size
onnx_file, // source
model_file // save to
);
}
auto yolo = Yolo::create_infer(model_file, Yolo::Type::X, 0, 0.4f);
if(yolo == nullptr){
INFOE("Engine is nullptr");
return;
}
//////////////////基础耗时////////////////////////
{
cv::VideoCapture cap("exp/face_tracker.mp4");
cv::Mat image;
int iframe = 0;
auto t0 = iLogger::timestamp_now_float();
while(iframe < 300 && cap.read(image)){
/** 模拟读取摄像头的延迟 **/
iLogger::sleep(40);
iframe++;
}
auto fee = iLogger::timestamp_now_float() - t0;
INFO("fee %.2f ms, fps = %.2f", fee, iframe / fee * 1000);
};
//////////////////传统做法////////////////////////
{
cv::VideoCapture cap("exp/face_tracker.mp4");
cv::Mat image;
int iframe = 0;
auto t0 = iLogger::timestamp_now_float();
while(iframe < 300 && cap.read(image)){
/** 模拟读取摄像头的延迟 **/
iLogger::sleep(40);
iframe++;
/** 立即拿结果，时序图效果差，耗时5.7ms **/
auto bboxes = yolo->commit(image).get();
//for(auto& box : bboxes)
// cv::rectangle(image, cv::Point(box.left, box.top), cv::Point(box.right, box.bottom), cv::Scalar(0, 255, 0), 2);
if(iframe % 100 == 0)
INFO("%d. %d objects", iframe++, bboxes.size());
}
auto fee = iLogger::timestamp_now_float() - t0;
INFO("fee %.2f ms, fps = %.2f", fee, iframe / fee * 1000);
};
//////////////////优化做法////////////////////////
{
cv::VideoCapture cap("exp/face_tracker.mp4");
shared_future<Yolo::ObjectBoxArray> prev_future;
cv::Mat image;
cv::Mat prev_image;
int iframe = 0;
auto t0 = iLogger::timestamp_now_float();
while(iframe < 300 && cap.read(image)){
/** 模拟读取摄像头的延迟 **/
iLogger::sleep(40);
iframe++;
if(prev_future.valid()){
auto bboxes = prev_future.get();
//for(auto& box : bboxes)
// cv::rectangle(prev_image, cv::Point(box.left, box.top), cv::Point(box.right, box.bottom), cv::Scalar(0, 255, 0), 2);
if(iframe % 100 == 0)
INFO("%d. %d objects", iframe++, bboxes.size());
}
image.copyTo(prev_image);
prev_future = yolo->commit(image);
}
auto fee = iLogger::timestamp_now_float() - t0;
INFO("fee %.2f ms, fps = %.2f", fee, iframe / fee * 1000);
};
}
int app_lesson(){
iLogger::set_log_level(iLogger::LogLevel::Verbose);
lesson1();
// lesson2();
// lesson3();
// lesson_cache1frame();
return 0;
}