Repository navigation
Expand file tree
/
Copy patharcface.cpp
More file actions
229 lines (191 loc) · 9.03 KB
/
Copy patharcface.cpp
File metadata and controls
229 lines (191 loc) · 9.03 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
#include "arcface.hpp"
#include <atomic>
#include <mutex>
#include <queue>
#include <condition_variable>
#include <infer/trt_infer.hpp>
#include <common/ilogger.hpp>
#include <common/infer_controller.hpp>
#include <common/preprocess_kernel.cuh>
#include <common/monopoly_allocator.hpp>
#include <common/cuda_tools.hpp>
namespace Arcface{
using namespace cv;
using namespace std;
struct AffineMatrix{
float i2d[6]; // image to dst(network), 2x3 matrix
float d2i[6];
void compute(const landmarks& lands){
// 112 x 112分辨率时的标准人脸关键点(训练用的是这个)
// 96 x 112分辨率时的标准人脸关键点在下面基础上去掉x的偏移
// 来源于论文和公开代码中训练用到的
// https://github.com/wy1iu/sphereface/blob/f5cd440a2233facf46b6529bd13231bb82f23177/preprocess/code/face_align_demo.m
float Sdata[] = {
30.2946 + 8, 51.6963,
65.5318 + 8, 51.5014,
48.0252 + 8, 71.7366,
33.5493 + 8, 92.3655,
62.7299 + 8, 92.2041
};
// 以下代码参考自:http://www.zifuture.com/archives/face-alignment
float Qdata[] = {
lands.points[0], lands.points[1], 1, 0,
lands.points[1], -lands.points[0], 0, 1,
lands.points[2], lands.points[3], 1, 0,
lands.points[3], -lands.points[2], 0, 1,
lands.points[4], lands.points[5], 1, 0,
lands.points[5], -lands.points[4], 0, 1,
lands.points[6], lands.points[7], 1, 0,
lands.points[7], -lands.points[6], 0, 1,
lands.points[8], lands.points[9], 1, 0,
lands.points[9], -lands.points[8], 0, 1,
};
float Udata[4];
Mat_<float> Q(10, 4, Qdata);
Mat_<float> U(4, 1, Udata);
Mat_<float> S(10, 1, Sdata);
U = (Q.t() * Q).inv() * Q.t() * S;
i2d[0] = Udata[0]; i2d[1] = Udata[1]; i2d[2] = Udata[2];
i2d[3] = -Udata[1]; i2d[4] = Udata[0]; i2d[5] = Udata[3];
cv::Mat m2x3_i2d(2, 3, CV_32F, i2d);
cv::Mat m2x3_d2i(2, 3, CV_32F, d2i);
cv::invertAffineTransform(m2x3_i2d, m2x3_d2i);
}
};
using ControllerImpl = InferController
<
commit_input, // input
feature, // output
tuple<string, int>, // start param
AffineMatrix // additional
>;
class InferImpl : public Infer, public ControllerImpl{
public:
/** 要求在InferImpl里面执行stop,而不是在基类执行stop **/
virtual ~InferImpl(){
stop();
}
virtual bool startup(const string& file, int gpuid){
float mean[] = {0.5f, 0.5f, 0.5f};
float std[] = {0.5f, 0.5f, 0.5f};
normalize_ = CUDAKernel::Norm::mean_std(mean, std, 1.0f / 255.0f, CUDAKernel::ChannelType::Invert);
return ControllerImpl::startup(make_tuple(file, gpuid));
}
virtual void worker(promise<bool>& result) override{
string file = get<0>(start_param_);
int gpuid = get<1>(start_param_);
TRT::set_device(gpuid);
auto engine = TRT::load_infer(file);
if(engine == nullptr){
INFOE("Engine %s load failed", file.c_str());
result.set_value(false);
return;
}
engine->print();
int max_batch_size = engine->get_max_batch_size();
auto input = engine->input();
auto output = engine->output();
input_width_ = input->size(3);
input_height_ = input->size(2);
feature_length_ = output->size(1);
tensor_allocator_ = make_shared<MonopolyAllocator<TRT::Tensor>>(max_batch_size * 2);
stream_ = engine->get_stream();
gpu_ = gpuid;
result.set_value(true);
input->resize_single_dim(0, max_batch_size).to_gpu();
output->resize_single_dim(0, max_batch_size).to_gpu();
vector<Job> fetch_jobs;
while(get_jobs_and_wait(fetch_jobs, max_batch_size)){
int infer_batch_size = fetch_jobs.size();
input->resize_single_dim(0, infer_batch_size);
for(int ibatch = 0; ibatch < infer_batch_size; ++ibatch){
auto& job = fetch_jobs[ibatch];
auto& mono = job.mono_tensor->data();
input->copy_from_gpu(input->offset(ibatch), mono->gpu(), mono->count());
job.mono_tensor->release();
}
engine->forward(false);
CUDAKernel::norm_feature(output->gpu<float>(), output->size(0), output->size(1), stream_);
output->to_cpu();
for(int ibatch = 0; ibatch < infer_batch_size; ++ibatch){
auto& job = fetch_jobs[ibatch];
float* image_based_output = output->cpu<float>(ibatch);
memcpy(job.output.ptr<float>(0), image_based_output, sizeof(float) * feature_length_);
job.pro->set_value(job.output);
}
fetch_jobs.clear();
}
INFO("Engine destroy.");
}
virtual bool preprocess(Job& job, const commit_input& input) override{
job.mono_tensor = tensor_allocator_->query();
if(job.mono_tensor == nullptr){
INFOE("Tensor allocator query failed.");
return false;
}
CUDATools::AutoDevice auto_device(gpu_);
auto& tensor = job.mono_tensor->data();
if(tensor == nullptr){
// not init
tensor = make_shared<TRT::Tensor>();
tensor->set_workspace(make_shared<TRT::MixMemory>());
}
auto& image = get<0>(input);
Size input_size(input_width_, input_height_);
job.additional.compute(get<1>(input));
job.output = Mat_<float>(1, feature_length_);
tensor->set_stream(stream_);
tensor->resize(1, 3, input_height_, input_width_);
size_t size_image = image.cols * image.rows * 3;
size_t size_matrix = iLogger::upbound(sizeof(job.additional.d2i), 32);
auto workspace = tensor->get_workspace();
uint8_t* gpu_workspace = (uint8_t*)workspace->gpu(size_matrix + size_image);
float* affine_matrix_device = (float*)gpu_workspace;
uint8_t* image_device = size_matrix + gpu_workspace;
uint8_t* cpu_workspace = (uint8_t*)workspace->cpu(size_matrix + size_image);
float* affine_matrix_host = (float*)cpu_workspace;
uint8_t* image_host = size_matrix + cpu_workspace;
//checkCudaRuntime(cudaMemcpyAsync(image_host, image.data, size_image, cudaMemcpyHostToHost, stream_));
// speed up
memcpy(image_host, image.data, size_image);
memcpy(affine_matrix_host, job.additional.d2i, sizeof(job.additional.d2i));
checkCudaRuntime(cudaMemcpyAsync(image_device, image_host, size_image, cudaMemcpyHostToDevice, stream_));
checkCudaRuntime(cudaMemcpyAsync(affine_matrix_device, affine_matrix_host, sizeof(job.additional.d2i), cudaMemcpyHostToDevice, stream_));
CUDAKernel::warp_affine_bilinear_and_normalize_plane(
image_device, image.cols * 3, image.cols, image.rows,
tensor->gpu<float>(), input_width_, input_height_,
affine_matrix_device, 0,
normalize_, stream_
);
return true;
}
virtual vector<shared_future<feature>> commits(const vector<commit_input>& inputs) override{
return ControllerImpl::commits(inputs);
}
virtual std::shared_future<feature> commit(const commit_input& input) override{
return ControllerImpl::commit(input);
}
private:
int input_width_ = 0;
int input_height_ = 0;
int feature_length_ = 0;
int gpu_ = 0;
TRT::CUStream stream_ = nullptr;
CUDAKernel::Norm normalize_;
};
Mat face_alignment(const cv::Mat& image, const landmarks& landmark){
Size input_size(112, 112);
AffineMatrix am;
am.compute(landmark);
Mat output;
warpAffine(image, output, Mat_<float>(2, 3, am.i2d), input_size, cv::INTER_LINEAR);
return output;
}
shared_ptr<Infer> create_infer(const string& engine_file, int gpuid){
shared_ptr<InferImpl> instance(new InferImpl());
if(!instance->startup(engine_file, gpuid)){
instance.reset();
}
return instance;
}
};