forked from sophon-ai-algo/examples
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathinference3.h
275 lines (206 loc) · 8.62 KB
/
inference3.h
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
271
272
273
274
275
#ifndef INFERENCE_FRAMEWORK_INFERENCE3_H
#define INFERENCE_FRAMEWORK_INFERENCE3_H
#include "bmutility.h"
#include "thread_queue.h"
namespace bm {
template<typename T1, typename T2>
class MediaDelegate {
public:
virtual ~MediaDelegate() {}
virtual int draw(std::vector<T2>& frames, std::vector<T2>& output) = 0;
virtual int stitch(std::vector<T2>& frames, std::vector<T1>& output) = 0;
virtual int encode(std::vector<T1>& frames) = 0;
};
struct MediaParam {
MediaParam() {}
int draw_queue_size;
int draw_thread_num;
int stitch_queue_size;
int stitch_thread_num;
int encode_queue_size;
int encode_thread_num;
bool stitch_blocking_push;
bool encode_blocking_push;
};
// for one thread mode
template<typename T1, typename T2>
class BMMediaPipeline {
std::shared_ptr<MediaDelegate<T1, T2>> m_media_delegate;
std::shared_ptr<BlockingQueue<T2>> m_drawQue;
WorkerPool<T2> m_drawWorkerPool;
std::shared_ptr<BlockingQueue<T2>> m_stitchQue;
WorkerPool<T2> m_stitchWorkerPool;
std::shared_ptr<BlockingQueue<T1>> m_encodeQue;
WorkerPool<T1> m_encodeWorkerPool;
public:
BMMediaPipeline() {
}
virtual ~BMMediaPipeline() {
}
void flow_control(std::function<void(T2& obj)> fn1,
std::function<void(T2& obj)> fn2,
std::function<void(T1& obj)> fn3) {
m_drawQue->set_drop_fn(fn1);
m_stitchQue->set_drop_fn(fn2);
m_encodeQue->set_drop_fn(fn3);
}
template<typename... T>
int init(const MediaParam ¶m, std::shared_ptr<MediaDelegate<T1, T2>> delegate, T&&... args) {
m_media_delegate = delegate;
const int underlying_type_std_queue = 0;
m_drawQue = std::make_shared<BlockingQueue<T2>>(
"draw", underlying_type_std_queue,
param.draw_queue_size);
m_stitchQue = std::make_shared<BlockingQueue<T2>>(
"stitch", underlying_type_std_queue,
param.stitch_queue_size);
m_drawWorkerPool.init(m_drawQue.get(), 1, 1, 1);
m_drawWorkerPool.startWork([this, ¶m](std::vector<T2> &items) {
std::vector<T2> frames;
m_media_delegate->draw(items, frames);
this->m_stitchQue->push(frames);
});
m_stitchWorkerPool.init(m_stitchQue.get(), 1, 1, 1);
m_stitchWorkerPool.startWork([this, ¶m](std::vector<T2> &items) {
std::vector<T1> frames;
m_media_delegate->stitch(items, frames);
this->m_encodeQue->push(frames);
});
m_encodeQue = std::make_shared<BlockingQueue<T1>>(
"encode", underlying_type_std_queue,
param.encode_queue_size);
m_encodeWorkerPool.init(m_encodeQue.get(), 1, 1, 8);
m_encodeWorkerPool.startWork([this, ¶m](std::vector<T1> &items) {
m_media_delegate->encode(items);
});
flow_control(std::forward<T>(args)...);
return 0;
}
int flush_frame() {
m_drawWorkerPool.flush();
return 0;
}
int push_frame(T2& frame) {
m_drawQue->push(frame);
return 0;
}
int push_frame(T1 *frame) {
m_drawQue->push(*frame);
return 0;
}
};
template<typename T1, typename T2>
class DetectorDelegate {
protected:
using DetectedFinishFunc = std::function<void(T2 &of)>;
DetectedFinishFunc m_pfnDetectFinish;
BMMediaPipeline<T1, T2>* m_nextMediaPipe = nullptr;
public:
virtual ~DetectorDelegate() {}
void set_next_inference_pipe(BMMediaPipeline<T1, T2> *nextPipe) { m_nextMediaPipe = nextPipe; }
virtual void decode_process(T1 &) {
// do nothing by default
}
virtual int preprocess(std::vector<T1> &frames, std::vector<T2> &of) = 0;
virtual int forward(std::vector<T2> &frames) = 0;
virtual int postprocess(std::vector<T2> &frames) = 0;
virtual int track(std::vector<T2> &frames) = 0;
virtual int set_detected_callback(DetectedFinishFunc func) { m_pfnDetectFinish = func; return 0;};
};
struct DetectorParam {
DetectorParam() {
preprocess_queue_size = 5;
preprocess_thread_num = 4;
inference_queue_size = 5;
inference_thread_num = 1;
postprocess_queue_size = 5;
postprocess_thread_num = 2;
batch_num=4;
}
int preprocess_queue_size;
int preprocess_thread_num;
int inference_queue_size;
int inference_thread_num;
int postprocess_queue_size;
int postprocess_thread_num;
int track_queue_size;
int track_thread_num;
int batch_num;
};
template<typename T1, typename T2>
class BMInferencePipe {
DetectorParam m_param;
std::shared_ptr<DetectorDelegate<T1, T2>> m_detect_delegate;
std::shared_ptr<BlockingQueue<T1>> m_preprocessQue;
std::shared_ptr<BlockingQueue<T2>> m_postprocessQue;
std::shared_ptr<BlockingQueue<T2>> m_forwardQue;
std::shared_ptr<BlockingQueue<T2>> m_trackQue;
WorkerPool<T1> m_preprocessWorkerPool;
WorkerPool<T2> m_forwardWorkerPool;
WorkerPool<T2> m_postprocessWorkerPool;
WorkerPool<T2> m_trackWorkerPool;
public:
BMInferencePipe() {
}
virtual ~BMInferencePipe() {
}
void flow_control(std::function<void(T1& obj)> fn1,
std::function<void(T2& obj)> fn2,
std::function<void(T2& obj)> fn3,
std::function<void(T2& obj)> fn4) {
m_preprocessQue->set_drop_fn(fn1);
m_forwardQue->set_drop_fn(fn2);
m_postprocessQue->set_drop_fn(fn3);
m_trackQue->set_drop_fn(fn4);
}
template<typename... T>
int init(const DetectorParam ¶m, std::shared_ptr<DetectorDelegate<T1, T2>> delegate, T&&... args) {
m_param = param;
m_detect_delegate = delegate;
const int underlying_type_std_queue = 0;
m_preprocessQue = std::make_shared<BlockingQueue<T1>>(
"preprocess", underlying_type_std_queue,
param.preprocess_queue_size);
m_postprocessQue = std::make_shared<BlockingQueue<T2>>(
"postprocess", underlying_type_std_queue,
param.postprocess_queue_size);
m_trackQue = std::make_shared<BlockingQueue<T2>>(
"track", underlying_type_std_queue,
param.track_queue_size);
m_forwardQue = std::make_shared<BlockingQueue<T2>>(
"inference", underlying_type_std_queue,
param.inference_queue_size);
m_preprocessWorkerPool.init(m_preprocessQue.get(), param.preprocess_thread_num, param.batch_num, param.batch_num);
m_preprocessWorkerPool.startWork([this, ¶m](std::vector<T1> &items) {
std::vector<T2> frames;
m_detect_delegate->preprocess(items, frames);
this->m_forwardQue->push(frames);
});
m_forwardWorkerPool.init(m_forwardQue.get(), param.inference_thread_num, 1, 8);
m_forwardWorkerPool.startWork([this, ¶m](std::vector<T2> &items) {
m_detect_delegate->forward(items);
this->m_postprocessQue->push(items);
});
m_postprocessWorkerPool.init(m_postprocessQue.get(), param.postprocess_thread_num, 1, 8);
m_postprocessWorkerPool.startWork([this, ¶m](std::vector<T2> &items) {
m_detect_delegate->postprocess(items);
this->m_trackQue->push(items);
});
m_trackWorkerPool.init(m_trackQue.get(), param.track_thread_num, 1, 8);
m_trackWorkerPool.startWork([this, ¶m](std::vector<T2> &items) {
m_detect_delegate->track(items);
});
flow_control(std::forward<T>(args)...);
return 0;
}
int flush_frame() {
m_preprocessWorkerPool.flush();
return 0;
}
int push_frame(T1 *frame) {
m_preprocessQue->push(*frame);
return 0;
}
};
} // end namespace bm
#endif //INFERENCE_FRAMEWORK_INFERENCE3_H