takeaway_member_cls.cpp
12.3 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
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
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
/*
* @Author: yangzilong
* @Date: 2021-11-25 10:42:54
* @Last Modified by: yangzilong
* @Email: yangzilong@objecteye.com
* @Description:
*/
#include <algorithm>
#include "takeaway_member_cls.hpp"
#include "../reprocessing_module/CropImg.h"
namespace ai_engine_module
{
namespace takeaway_member_classification
{
algorithm_type_t TakeawayMemberCls::algor_type_ = algorithm_type_t::TAKEAWAY_MEMBER_CLASSIFICATION;
using namespace classification;
TakeawayMemberCls::TakeawayMemberCls()
: task_param_manager_(nullptr)
{
}
TakeawayMemberCls::~TakeawayMemberCls()
{
takeway_member::release(&tools_);
if (!tools_)
{
delete tools_;
tools_ = nullptr;
}
}
bool TakeawayMemberCls::init(int gpu_id, char* trt_serialize_file)
{
init_ = false;
common::init_params_t param;
{
param.mode = gpu_id >= 0 ? DEVICE_GPU : DEVICE_CPU;
param.gpuid = gpu_id;
param.threshold = 0.0;
param.engine = ENGINE_TENSORRT;
param.max_batch = MAX_BATCH;
param.trt_serialize_file = trt_serialize_file;
}
// helpers::os::mkdirp(trt_serialize_file);
int status;
if (!(init_ = (0 == (status = takeway_member::init(&tools_, ¶m)))))
LOG_ERROR("Init TakeawayMemberClsSdk failed error code is {}", status);
else
if (!task_param_manager_)
task_param_manager_ = task_param_manager::getInstance();
return init_;
}
bool TakeawayMemberCls::check_initied()
{
if (!init_)
LOG_ERROR("[%s:%d] call init function please.", __FILE__, __LINE__);
return init_;
}
void TakeawayMemberCls::force_release_result(const task_id_t& task_id) {
for (auto iter = id_to_result_.begin(); iter != id_to_result_.end();) {
const auto& key = iter->first;
if (key.task_id == task_id) {
auto& value = iter->second;
if (value.roi_img.data_ != nullptr) {
CHECK(cudaFree(value.roi_img.data_));
value.roi_img.data_ = nullptr;
}
if (value.ori_img.data_ != nullptr) {
CHECK(cudaFree(value.ori_img.data_));
value.ori_img.data_ = nullptr;
}
iter = id_to_result_.erase(iter);
}
else {
++iter;
}
}
}
std::shared_ptr<result_data_t> TakeawayMemberCls::get_result_by_objectid(const id_t& id, bool do_erase)
{
auto it = id_to_result_.find(id);
if (it == id_to_result_.end())
return std::shared_ptr<result_data_t>(nullptr);
std::shared_ptr<result_data_t> res = std::make_shared<result_data_t>(it->second);
if (do_erase)
id_to_result_.erase(id);
return res;
}
bool TakeawayMemberCls::update_mstreams(const std::set<task_id_t>& taskIds, const sy_img* det_input_images, const std::vector<onelevel_det_result>& det_results)
{
if (!check_initied())
return false;
if (det_results.empty())
{
LOG_DEBUG("detection result is empty.");
return false;
}
int n_images = det_results.size(); // or n_stream
unsigned flattened_idx = 0;
std::map<int, int> flattened_idx_to_batch_idx;
/* 1. Crop & keep some interest class. */
auto taskId_iter = taskIds.begin();
std::vector<sy_img> flattened_imgs(0);
std::vector<input_data_wrap_t> flattened_interest_data(0); //
for (int n = 0; n < n_images; ++n)
{
int n_interest_obj = 0;
auto& src_img = det_input_images[n];
auto& boxes_of_one_image = det_results[n].obj;
for (int i = 0; i < det_results[n].obj_count; ++i)
{
auto& box = boxes_of_one_image[i];
if (static_cast<det_class_label_t>(box.index) == det_class_label_t::MOTOCYCLE)
{
auto& taskId = *taskId_iter;
auto algor_param_wrap = task_param_manager_->get_task_other_param(taskId, this->algor_type_);
if (!algor_param_wrap)
{
LOG_ERROR("{} is nullptr when get algor param from task_param", taskId.c_str());
continue;
}
auto algor_param = ((algor_param_type)algor_param_wrap->algor_param);
input_data_wrap_t data;
int top = std::max(int(box.top - (IMAGE_CROP_EXPAND_RATIO * box.top)), 0);
int left = std::max(int(box.left - (IMAGE_CROP_EXPAND_RATIO * box.left)), 0);
int right = std::min(int(box.right + (IMAGE_CROP_EXPAND_RATIO * box.right)), src_img.w_);
int bottom = std::min(int(box.bottom + (IMAGE_CROP_EXPAND_RATIO * box.bottom)), src_img.h_);
int width = right - left;
int height = bottom - top;
if ((width < algor_param->pedestrian_min_width || height < algor_param->pedestrian_min_height || box.confidence < algor_param->pedestrian_confidence_threshold) ||
!snapshot_legal_inarea(algor_param_wrap->basic_param->algor_valid_rect, left, top, right, bottom))
continue;
data.box.top = top;
data.box.left = left;
data.box.right = right;
data.box.bottom = bottom;
data.taskId = taskId;
data.objId = box.id;
data.id = obj_key_t{ box.id, taskId, algorithm_type_t::TAKEAWAY_MEMBER_CLASSIFICATION };
sy_img img;
{
img.w_ = width;
img.h_ = height;
img.c_ = src_img.c_;
}
cudaError_t cuda_status;
const unsigned nbytes = img.c_ * img.h_ * img.w_ * sizeof(unsigned char);
if (CUDA_SUCCESS != (cuda_status = cudaMalloc((void**)&img.data_, nbytes)))
{
LOG_ERROR("cudaMalloc failed: {} malloc nbytes is {} mb is {} ", cudaGetErrorString(cuda_status), nbytes, nbytes / (1024 * 1024));
continue;
}
if (CUDA_SUCCESS != (cuda_status = cudacommon::CropImgGpu(src_img.data_, src_img.w_, src_img.h_, img.data_, left, top, width, height)))
{
LOG_ERROR("Crop image GPU failed error is {} wh is [{}, {}] ltrb is [{} {} {} {}]",
cudaGetErrorString(cuda_status), src_img.w_, src_img.h_, data.box.left, data.box.top, data.box.right, data.box.bottom);
CHECK(cudaFree(img.data_));
continue;
}
flattened_imgs.emplace_back(std::move(img));
flattened_interest_data.emplace_back(std::move(data));
flattened_idx_to_batch_idx[flattened_idx++] = n;
}
}
++taskId_iter;
}
/* 2. collection result. */
int n_input_image = flattened_imgs.size();
takeway_member::results_t model_results[n_input_image];
{
int steps = (n_input_image + MAX_BATCH - 1) / MAX_BATCH;
for (int step = 0; step < steps; ++step)
{
int offset = step * MAX_BATCH;
int batch_size = (step == steps - 1) ? n_input_image - offset : MAX_BATCH;
takeway_member::process_batch(tools_, flattened_imgs.data() + offset, batch_size, model_results + offset);
}
}
/* 3. postprocess. */
{
for (int n = 0; n < n_input_image; ++n)
{
auto& det_result = flattened_interest_data[n];
auto& objId = det_result.objId;
if (id_to_result_.find(det_result.id) != id_to_result_.end())
{
CHECK(cudaFree(flattened_imgs[n].data_));
flattened_imgs[n].data_ = nullptr;
continue;
}
const sy_img& src_img = det_input_images[flattened_idx_to_batch_idx[n]];
auto algor_param_wrap = task_param_manager_->get_task_other_param(det_result.taskId, this->algor_type_);
if (!algor_param_wrap)
{
LOG_ERROR("{} nullptr when get algor param from task_param", det_result.taskId.c_str());
CHECK(cudaFree(flattened_imgs[n].data_));
flattened_imgs[n].data_ = nullptr;
continue;
}
auto algor_param = ((algor_param_type)algor_param_wrap->algor_param);
takeway_member::takeaway_member_label_t takeway_member_cls = static_cast<takeway_member::takeaway_member_label_t>(
model_results[n].multi_label_cls_result[(int)takeway_member::label_index_t::TAKEAWAY_MEMBER].category);
det_result.box.score = model_results[n].multi_label_cls_result[(int)takeway_member::label_index_t::TAKEAWAY_MEMBER].prob;
obj_key_t obj_key{ det_result.objId, det_result.taskId, algorithm_type_t::TAKEAWAY_MEMBER_CLASSIFICATION };
auto& e = id_to_mn_[obj_key];
++e.m_frame;
if (takeway_member_cls != takeway_member::takeaway_member_label_t::NOT &&
det_result.box.score >= algor_param->threshold)
{
if (++e.n_frame == algor_param->n)
{
result_data_t result;
{
result.box = det_result.box;
result.taskId = det_result.taskId;
result.objId = det_result.objId;
#if 0
{
result.ori_img = src_img;
}
#else
{
sy_img img;
{
img.c_ = src_img.c_;
img.h_ = src_img.h_;
img.w_ = src_img.w_;
}
unsigned nbytes = img.c_ * img.h_ * img.w_ * sizeof(unsigned char);
CHECK(cudaMalloc(&img.data_, nbytes));
CHECK(cudaMemcpy(img.data_, src_img.data_, nbytes, cudaMemcpyDeviceToDevice));
result.ori_img = std::move(img);
}
#endif
result.roi_img = std::move(flattened_imgs[n]);
result.category = (int)takeway_member_cls;
}
id_to_result_.emplace(obj_key, std::move(result));
goto _continue;
}
}
if (e.m_frame == algor_param->m)
e.reset();
CHECK(cudaFree(flattened_imgs[n].data_));
_continue:
{
}
}
}
return true;
}
} // namespace takeaway_member_clasiication
} // namespace ai_engine_module