提交 944ffc57 编写于 作者: D dongshuilong

code format for cls cpp infer

上级 4dbd85b8
...@@ -35,55 +35,55 @@ using namespace paddle_infer; ...@@ -35,55 +35,55 @@ using namespace paddle_infer;
namespace PaddleClas { namespace PaddleClas {
class Classifier { class Classifier {
public: public:
explicit Classifier(const ClsConfig &config) { explicit Classifier(const ClsConfig &config) {
this->use_gpu_ = config.use_gpu; this->use_gpu_ = config.use_gpu;
this->gpu_id_ = config.gpu_id; this->gpu_id_ = config.gpu_id;
this->gpu_mem_ = config.gpu_mem; this->gpu_mem_ = config.gpu_mem;
this->cpu_math_library_num_threads_ = config.cpu_threads; this->cpu_math_library_num_threads_ = config.cpu_threads;
this->use_fp16_ = config.use_fp16; this->use_fp16_ = config.use_fp16;
this->use_mkldnn_ = config.use_mkldnn; this->use_mkldnn_ = config.use_mkldnn;
this->use_tensorrt_ = config.use_tensorrt; this->use_tensorrt_ = config.use_tensorrt;
this->mean_ = config.mean; this->mean_ = config.mean;
this->std_ = config.std; this->std_ = config.std;
this->resize_short_size_ = config.resize_short_size; this->resize_short_size_ = config.resize_short_size;
this->scale_ = config.scale; this->scale_ = config.scale;
this->crop_size_ = config.crop_size; this->crop_size_ = config.crop_size;
this->ir_optim_ = config.ir_optim; this->ir_optim_ = config.ir_optim;
LoadModel(config.cls_model_path, config.cls_params_path); LoadModel(config.cls_model_path, config.cls_params_path);
} }
// Load Paddle inference model // Load Paddle inference model
void LoadModel(const std::string &model_path, const std::string &params_path); void LoadModel(const std::string &model_path, const std::string &params_path);
// Run predictor // Run predictor
double Run(cv::Mat &img, std::vector<double> *times); double Run(cv::Mat &img, std::vector<double> *times);
private: private:
std::shared_ptr<Predictor> predictor_; std::shared_ptr <Predictor> predictor_;
bool use_gpu_ = false; bool use_gpu_ = false;
int gpu_id_ = 0; int gpu_id_ = 0;
int gpu_mem_ = 4000; int gpu_mem_ = 4000;
int cpu_math_library_num_threads_ = 4; int cpu_math_library_num_threads_ = 4;
bool use_mkldnn_ = false; bool use_mkldnn_ = false;
bool use_tensorrt_ = false; bool use_tensorrt_ = false;
bool use_fp16_ = false; bool use_fp16_ = false;
bool ir_optim_ = true; bool ir_optim_ = true;
std::vector<float> mean_ = {0.485f, 0.456f, 0.406f}; std::vector<float> mean_ = {0.485f, 0.456f, 0.406f};
std::vector<float> std_ = {0.229f, 0.224f, 0.225f}; std::vector<float> std_ = {0.229f, 0.224f, 0.225f};
float scale_ = 0.00392157; float scale_ = 0.00392157;
int resize_short_size_ = 256; int resize_short_size_ = 256;
int crop_size_ = 224; int crop_size_ = 224;
// pre-process // pre-process
ResizeImg resize_op_; ResizeImg resize_op_;
Normalize normalize_op_; Normalize normalize_op_;
Permute permute_op_; Permute permute_op_;
CenterCropImg crop_op_; CenterCropImg crop_op_;
}; };
} // namespace PaddleClas } // namespace PaddleClas
...@@ -31,81 +31,83 @@ ...@@ -31,81 +31,83 @@
namespace PaddleClas { namespace PaddleClas {
class ClsConfig { class ClsConfig {
public: public:
explicit ClsConfig(const std::string &path) { explicit ClsConfig(const std::string &path) {
ReadYamlConfig(path); ReadYamlConfig(path);
this->infer_imgs = this->infer_imgs =
this->config_file["Global"]["infer_imgs"].as<std::string>(); this->config_file["Global"]["infer_imgs"].as<std::string>();
this->batch_size = this->config_file["Global"]["batch_size"].as<int>(); this->batch_size = this->config_file["Global"]["batch_size"].as<int>();
this->use_gpu = this->config_file["Global"]["use_gpu"].as<bool>(); this->use_gpu = this->config_file["Global"]["use_gpu"].as<bool>();
if (this->config_file["Global"]["gpu_id"].IsDefined()) if (this->config_file["Global"]["gpu_id"].IsDefined())
this->gpu_id = this->config_file["Global"]["gpu_id"].as<int>(); this->gpu_id = this->config_file["Global"]["gpu_id"].as<int>();
else else
this->gpu_id = 0; this->gpu_id = 0;
this->gpu_mem = this->config_file["Global"]["gpu_mem"].as<int>(); this->gpu_mem = this->config_file["Global"]["gpu_mem"].as<int>();
this->cpu_threads = this->cpu_threads =
this->config_file["Global"]["cpu_num_threads"].as<int>(); this->config_file["Global"]["cpu_num_threads"].as<int>();
this->use_mkldnn = this->config_file["Global"]["enable_mkldnn"].as<bool>(); this->use_mkldnn = this->config_file["Global"]["enable_mkldnn"].as<bool>();
this->use_tensorrt = this->config_file["Global"]["use_tensorrt"].as<bool>(); this->use_tensorrt = this->config_file["Global"]["use_tensorrt"].as<bool>();
this->use_fp16 = this->config_file["Global"]["use_fp16"].as<bool>(); this->use_fp16 = this->config_file["Global"]["use_fp16"].as<bool>();
this->enable_benchmark = this->enable_benchmark =
this->config_file["Global"]["enable_benchmark"].as<bool>(); this->config_file["Global"]["enable_benchmark"].as<bool>();
this->ir_optim = this->config_file["Global"]["ir_optim"].as<bool>(); this->ir_optim = this->config_file["Global"]["ir_optim"].as<bool>();
this->enable_profile = this->enable_profile =
this->config_file["Global"]["enable_profile"].as<bool>(); this->config_file["Global"]["enable_profile"].as<bool>();
this->cls_model_path = this->cls_model_path =
this->config_file["Global"]["inference_model_dir"].as<std::string>() + this->config_file["Global"]["inference_model_dir"].as<std::string>() +
OS_PATH_SEP + "inference.pdmodel"; OS_PATH_SEP + "inference.pdmodel";
this->cls_params_path = this->cls_params_path =
this->config_file["Global"]["inference_model_dir"].as<std::string>() + this->config_file["Global"]["inference_model_dir"].as<std::string>() +
OS_PATH_SEP + "inference.pdiparams"; OS_PATH_SEP + "inference.pdiparams";
this->resize_short_size = this->resize_short_size =
this->config_file["PreProcess"]["transform_ops"][0]["ResizeImage"] this->config_file["PreProcess"]["transform_ops"][0]["ResizeImage"]
["resize_short"] ["resize_short"]
.as<int>(); .as<int>();
this->crop_size = this->crop_size =
this->config_file["PreProcess"]["transform_ops"][1]["CropImage"]["size"] this->config_file["PreProcess"]["transform_ops"][1]["CropImage"]["size"]
.as<int>(); .as<int>();
this->scale = this->config_file["PreProcess"]["transform_ops"][2] this->scale = this->config_file["PreProcess"]["transform_ops"][2]
["NormalizeImage"]["scale"] ["NormalizeImage"]["scale"]
.as<float>(); .as<float>();
this->mean = this->config_file["PreProcess"]["transform_ops"][2] this->mean = this->config_file["PreProcess"]["transform_ops"][2]
["NormalizeImage"]["mean"] ["NormalizeImage"]["mean"]
.as<std::vector<float>>(); .as < std::vector < float >> ();
this->std = this->config_file["PreProcess"]["transform_ops"][2] this->std = this->config_file["PreProcess"]["transform_ops"][2]
["NormalizeImage"]["std"] ["NormalizeImage"]["std"]
.as<std::vector<float>>(); .as < std::vector < float >> ();
if (this->config_file["Global"]["benchmark"].IsDefined()) if (this->config_file["Global"]["benchmark"].IsDefined())
this->benchmark = this->config_file["Global"]["benchmark"].as<bool>(); this->benchmark = this->config_file["Global"]["benchmark"].as<bool>();
else else
this->benchmark = false; this->benchmark = false;
} }
YAML::Node config_file; YAML::Node config_file;
bool use_gpu = false; bool use_gpu = false;
int gpu_id = 0; int gpu_id = 0;
int gpu_mem = 4000; int gpu_mem = 4000;
int cpu_threads = 1; int cpu_threads = 1;
bool use_mkldnn = false; bool use_mkldnn = false;
bool use_tensorrt = false; bool use_tensorrt = false;
bool use_fp16 = false; bool use_fp16 = false;
bool benchmark = false; bool benchmark = false;
int batch_size = 1; int batch_size = 1;
bool enable_benchmark = false; bool enable_benchmark = false;
bool ir_optim = true; bool ir_optim = true;
bool enable_profile = false; bool enable_profile = false;
std::string cls_model_path; std::string cls_model_path;
std::string cls_params_path; std::string cls_params_path;
std::string infer_imgs; std::string infer_imgs;
int resize_short_size = 256; int resize_short_size = 256;
int crop_size = 224; int crop_size = 224;
float scale = 0.00392157; float scale = 0.00392157;
std::vector<float> mean = {0.485, 0.456, 0.406}; std::vector<float> mean = {0.485, 0.456, 0.406};
std::vector<float> std = {0.229, 0.224, 0.225}; std::vector<float> std = {0.229, 0.224, 0.225};
void PrintConfigInfo();
void ReadYamlConfig(const std::string &path); void PrintConfigInfo();
};
void ReadYamlConfig(const std::string &path);
};
} // namespace PaddleClas } // namespace PaddleClas
...@@ -31,26 +31,26 @@ using namespace std; ...@@ -31,26 +31,26 @@ using namespace std;
namespace PaddleClas { namespace PaddleClas {
class Normalize { class Normalize {
public: public:
virtual void Run(cv::Mat *im, const std::vector<float> &mean, virtual void Run(cv::Mat *im, const std::vector<float> &mean,
const std::vector<float> &std, float &scale); const std::vector<float> &std, float &scale);
}; };
// RGB -> CHW // RGB -> CHW
class Permute { class Permute {
public: public:
virtual void Run(const cv::Mat *im, float *data); virtual void Run(const cv::Mat *im, float *data);
}; };
class CenterCropImg { class CenterCropImg {
public: public:
virtual void Run(cv::Mat &im, const int crop_size = 224); virtual void Run(cv::Mat &im, const int crop_size = 224);
}; };
class ResizeImg { class ResizeImg {
public: public:
virtual void Run(const cv::Mat &img, cv::Mat &resize_img, int max_size_len); virtual void Run(const cv::Mat &img, cv::Mat &resize_img, int max_size_len);
}; };
} // namespace PaddleClas } // namespace PaddleClas
...@@ -32,15 +32,15 @@ ...@@ -32,15 +32,15 @@
namespace PaddleClas { namespace PaddleClas {
class Utility { class Utility {
public: public:
static std::vector<std::string> ReadDict(const std::string &path); static std::vector <std::string> ReadDict(const std::string &path);
// template <class ForwardIterator> // template <class ForwardIterator>
// inline static size_t argmax(ForwardIterator first, ForwardIterator last) // inline static size_t argmax(ForwardIterator first, ForwardIterator last)
// { // {
// return std::distance(first, std::max_element(first, last)); // return std::distance(first, std::max_element(first, last));
// } // }
}; };
} // namespace PaddleClas } // namespace PaddleClas
\ No newline at end of file
...@@ -143,10 +143,10 @@ tar -xvf paddle_inference.tgz ...@@ -143,10 +143,10 @@ tar -xvf paddle_inference.tgz
``` ```
inference/ inference/
|--cls_infer.pdmodel |--inference.pdmodel
|--cls_infer.pdiparams |--inference.pdiparams
``` ```
**注意**:上述文件中,`cls_infer.pdmodel`文件存储了模型结构信息,`cls_infer.pdiparams`文件存储了模型参数信息。注意两个文件的路径需要与配置文件`tools/config.txt`中的`cls_model_path``cls_params_path`参数对应一致 **注意**:上述文件中,`inference.pdmodel`文件存储了模型结构信息,`inference.pdiparams`文件存储了模型参数信息。模型目录可以随意设置,但是模型名字不能修改
### 2.2 编译PaddleClas C++预测demo ### 2.2 编译PaddleClas C++预测demo
...@@ -183,6 +183,7 @@ cmake .. \ ...@@ -183,6 +183,7 @@ cmake .. \
-DCUDA_LIB=${CUDA_LIB_DIR} \ -DCUDA_LIB=${CUDA_LIB_DIR} \
make -j make -j
cd ..
``` ```
上述命令中, 上述命令中,
...@@ -200,31 +201,26 @@ make -j ...@@ -200,31 +201,26 @@ make -j
在执行上述命令,编译完成之后,会在当前路径下生成`build`文件夹,其中生成一个名为`clas_system`的可执行文件。 在执行上述命令,编译完成之后,会在当前路径下生成`build`文件夹,其中生成一个名为`clas_system`的可执行文件。
### 运行demo ### 2.3 运行demo
* 首先修改`tools/config.txt`中对应字段: #### 2.3.1 设置配置文件
* use_gpu:是否使用GPU;
* gpu_id:使用的GPU卡号;
* gpu_mem:显存;
* cpu_math_library_num_threads:底层科学计算库所用线程的数量;
* use_mkldnn:是否使用MKLDNN加速;
* use_tensorrt: 是否使用tensorRT进行加速;
* use_fp16:是否使用半精度浮点数进行计算,该选项仅在use_tensorrt为true时有效;
* cls_model_path:预测模型结构文件路径;
* cls_params_path:预测模型参数文件路径;
* resize_short_size:预处理时图像缩放大小;
* crop_size:预处理时图像裁剪后的大小。
* 然后修改`tools/run.sh` ```shell
* `./build/clas_system ./tools/config.txt ./docs/imgs/ILSVRC2012_val_00000666.JPEG` cp ../configs/inference_cls.yaml tools/
* 上述命令中分别为:编译得到的可执行文件`clas_system`;运行时的配置文件`config.txt`;待预测的图像。 ```
根据[python预测推理](../../docs/zh_CN/inference_deployment/python_deploy.md)`图像分类推理`部分修改好`tools`目录下`inference_cls.yaml`文件。`yaml`文件的参数说明详见[python预测推理](../../docs/zh_CN/inference_deployment/python_deploy.md)
请根据实际存放文件,修改好`Global.infer_imgs``Global.inference_model_dir`等参数。
* 最后执行以下命令,完成对一幅图像的分类。 #### 2.3.2 执行
```shell ```shell
sh tools/run.sh ./build/clas_system -c inference_cls.yaml
# or
./build/clas_system -config inference_cls.yaml
``` ```
* 最终屏幕上会输出结果,如下图所示。 最终屏幕上会输出结果,如下图所示。
<div align="center"> <div align="center">
<img src="./docs/imgs/cpp_infer_result.png" width="600"> <img src="./docs/imgs/cpp_infer_result.png" width="600">
......
...@@ -16,97 +16,97 @@ ...@@ -16,97 +16,97 @@
namespace PaddleClas { namespace PaddleClas {
void Classifier::LoadModel(const std::string &model_path, void Classifier::LoadModel(const std::string &model_path,
const std::string &params_path) { const std::string &params_path) {
paddle_infer::Config config; paddle_infer::Config config;
config.SetModel(model_path, params_path); config.SetModel(model_path, params_path);
if (this->use_gpu_) { if (this->use_gpu_) {
config.EnableUseGpu(this->gpu_mem_, this->gpu_id_); config.EnableUseGpu(this->gpu_mem_, this->gpu_id_);
if (this->use_tensorrt_) { if (this->use_tensorrt_) {
config.EnableTensorRtEngine( config.EnableTensorRtEngine(
1 << 20, 1, 3, 1 << 20, 1, 3,
this->use_fp16_ ? paddle_infer::Config::Precision::kHalf this->use_fp16_ ? paddle_infer::Config::Precision::kHalf
: paddle_infer::Config::Precision::kFloat32, : paddle_infer::Config::Precision::kFloat32,
false, false); false, false);
}
} else {
config.DisableGpu();
if (this->use_mkldnn_) {
config.EnableMKLDNN();
// cache 10 different shapes for mkldnn to avoid memory leak
config.SetMkldnnCacheCapacity(10);
}
config.SetCpuMathLibraryNumThreads(this->cpu_math_library_num_threads_);
}
config.SwitchUseFeedFetchOps(false);
// true for multiple input
config.SwitchSpecifyInputNames(true);
config.SwitchIrOptim(this->ir_optim_);
config.EnableMemoryOptim();
config.DisableGlogInfo();
this->predictor_ = CreatePredictor(config);
} }
} else {
config.DisableGpu(); double Classifier::Run(cv::Mat &img, std::vector<double> *times) {
if (this->use_mkldnn_) { cv::Mat srcimg;
config.EnableMKLDNN(); cv::Mat resize_img;
// cache 10 different shapes for mkldnn to avoid memory leak img.copyTo(srcimg);
config.SetMkldnnCacheCapacity(10);
auto preprocess_start = std::chrono::system_clock::now();
this->resize_op_.Run(img, resize_img, this->resize_short_size_);
this->crop_op_.Run(resize_img, this->crop_size_);
this->normalize_op_.Run(&resize_img, this->mean_, this->std_, this->scale_);
std::vector<float> input(1 * 3 * resize_img.rows * resize_img.cols, 0.0f);
this->permute_op_.Run(&resize_img, input.data());
auto input_names = this->predictor_->GetInputNames();
auto input_t = this->predictor_->GetInputHandle(input_names[0]);
input_t->Reshape({1, 3, resize_img.rows, resize_img.cols});
auto preprocess_end = std::chrono::system_clock::now();
auto infer_start = std::chrono::system_clock::now();
input_t->CopyFromCpu(input.data());
this->predictor_->Run();
std::vector<float> out_data;
auto output_names = this->predictor_->GetOutputNames();
auto output_t = this->predictor_->GetOutputHandle(output_names[0]);
std::vector<int> output_shape = output_t->shape();
int out_num = std::accumulate(output_shape.begin(), output_shape.end(), 1,
std::multiplies<int>());
out_data.resize(out_num);
output_t->CopyToCpu(out_data.data());
auto infer_end = std::chrono::system_clock::now();
auto postprocess_start = std::chrono::system_clock::now();
int maxPosition =
max_element(out_data.begin(), out_data.end()) - out_data.begin();
auto postprocess_end = std::chrono::system_clock::now();
std::chrono::duration<float> preprocess_diff =
preprocess_end - preprocess_start;
times->push_back(double(preprocess_diff.count() * 1000));
std::chrono::duration<float> inference_diff = infer_end - infer_start;
double inference_cost_time = double(inference_diff.count() * 1000);
times->push_back(inference_cost_time);
std::chrono::duration<float> postprocess_diff =
postprocess_end - postprocess_start;
times->push_back(double(postprocess_diff.count() * 1000));
std::cout << "result: " << std::endl;
std::cout << "\tclass id: " << maxPosition << std::endl;
std::cout << std::fixed << std::setprecision(10)
<< "\tscore: " << double(out_data[maxPosition]) << std::endl;
return inference_cost_time;
} }
config.SetCpuMathLibraryNumThreads(this->cpu_math_library_num_threads_);
}
config.SwitchUseFeedFetchOps(false);
// true for multiple input
config.SwitchSpecifyInputNames(true);
config.SwitchIrOptim(this->ir_optim_);
config.EnableMemoryOptim();
config.DisableGlogInfo();
this->predictor_ = CreatePredictor(config);
}
double Classifier::Run(cv::Mat &img, std::vector<double> *times) {
cv::Mat srcimg;
cv::Mat resize_img;
img.copyTo(srcimg);
auto preprocess_start = std::chrono::system_clock::now();
this->resize_op_.Run(img, resize_img, this->resize_short_size_);
this->crop_op_.Run(resize_img, this->crop_size_);
this->normalize_op_.Run(&resize_img, this->mean_, this->std_, this->scale_);
std::vector<float> input(1 * 3 * resize_img.rows * resize_img.cols, 0.0f);
this->permute_op_.Run(&resize_img, input.data());
auto input_names = this->predictor_->GetInputNames();
auto input_t = this->predictor_->GetInputHandle(input_names[0]);
input_t->Reshape({1, 3, resize_img.rows, resize_img.cols});
auto preprocess_end = std::chrono::system_clock::now();
auto infer_start = std::chrono::system_clock::now();
input_t->CopyFromCpu(input.data());
this->predictor_->Run();
std::vector<float> out_data;
auto output_names = this->predictor_->GetOutputNames();
auto output_t = this->predictor_->GetOutputHandle(output_names[0]);
std::vector<int> output_shape = output_t->shape();
int out_num = std::accumulate(output_shape.begin(), output_shape.end(), 1,
std::multiplies<int>());
out_data.resize(out_num);
output_t->CopyToCpu(out_data.data());
auto infer_end = std::chrono::system_clock::now();
auto postprocess_start = std::chrono::system_clock::now();
int maxPosition =
max_element(out_data.begin(), out_data.end()) - out_data.begin();
auto postprocess_end = std::chrono::system_clock::now();
std::chrono::duration<float> preprocess_diff =
preprocess_end - preprocess_start;
times->push_back(double(preprocess_diff.count() * 1000));
std::chrono::duration<float> inference_diff = infer_end - infer_start;
double inference_cost_time = double(inference_diff.count() * 1000);
times->push_back(inference_cost_time);
std::chrono::duration<float> postprocess_diff =
postprocess_end - postprocess_start;
times->push_back(double(postprocess_diff.count() * 1000));
std::cout << "result: " << std::endl;
std::cout << "\tclass id: " << maxPosition << std::endl;
std::cout << std::fixed << std::setprecision(10)
<< "\tscore: " << double(out_data[maxPosition]) << std::endl;
return inference_cost_time;
}
} // namespace PaddleClas } // namespace PaddleClas
...@@ -16,20 +16,20 @@ ...@@ -16,20 +16,20 @@
namespace PaddleClas { namespace PaddleClas {
void ClsConfig::PrintConfigInfo() { void ClsConfig::PrintConfigInfo() {
std::cout << "=======Paddle Class inference config======" << std::endl; std::cout << "=======Paddle Class inference config======" << std::endl;
std::cout << this->config_file << std::endl; std::cout << this->config_file << std::endl;
std::cout << "=======End of Paddle Class inference config======" << std::endl; std::cout << "=======End of Paddle Class inference config======" << std::endl;
} }
void ClsConfig::ReadYamlConfig(const std::string &path) { void ClsConfig::ReadYamlConfig(const std::string &path) {
try { try {
this->config_file = YAML::LoadFile(path); this->config_file = YAML::LoadFile(path);
} catch (YAML::BadFile &e) { } catch (YAML::BadFile &e) {
std::cout << "Something wrong in yaml file, please check yaml file" std::cout << "Something wrong in yaml file, please check yaml file"
<< std::endl; << std::endl;
exit(1); exit(1);
} }
} }
}; // namespace PaddleClas }; // namespace PaddleClas
...@@ -35,79 +35,81 @@ using namespace std; ...@@ -35,79 +35,81 @@ using namespace std;
using namespace cv; using namespace cv;
using namespace PaddleClas; using namespace PaddleClas;
DEFINE_string(config, "", "Path of yaml file"); DEFINE_string(config,
DEFINE_string(c, "", "Path of yaml file"); "", "Path of yaml file");
DEFINE_string(c,
"", "Path of yaml file");
int main(int argc, char **argv) { int main(int argc, char **argv) {
google::ParseCommandLineFlags(&argc, &argv, true); google::ParseCommandLineFlags(&argc, &argv, true);
std::string yaml_path = ""; std::string yaml_path = "";
if (FLAGS_config == "" && FLAGS_c == "") { if (FLAGS_config == "" && FLAGS_c == "") {
std::cerr << "[ERROR] usage: " << std::endl std::cerr << "[ERROR] usage: " << std::endl
<< argv[0] << " -c $yaml_path" << std::endl << argv[0] << " -c $yaml_path" << std::endl
<< "or:" << std::endl << "or:" << std::endl
<< argv[0] << " -config $yaml_path" << std::endl; << argv[0] << " -config $yaml_path" << std::endl;
exit(1); exit(1);
} else if (FLAGS_config != "") { } else if (FLAGS_config != "") {
yaml_path = FLAGS_config; yaml_path = FLAGS_config;
} else { } else {
yaml_path = FLAGS_c; yaml_path = FLAGS_c;
} }
ClsConfig config(yaml_path); ClsConfig config(yaml_path);
config.PrintConfigInfo(); config.PrintConfigInfo();
std::string path(config.infer_imgs); std::string path(config.infer_imgs);
std::vector<std::string> img_files_list; std::vector <std::string> img_files_list;
if (cv::utils::fs::isDirectory(path)) { if (cv::utils::fs::isDirectory(path)) {
std::vector<cv::String> filenames; std::vector <cv::String> filenames;
cv::glob(path, filenames); cv::glob(path, filenames);
for (auto f : filenames) { for (auto f : filenames) {
img_files_list.push_back(f); img_files_list.push_back(f);
}
} else {
img_files_list.push_back(path);
} }
} else {
img_files_list.push_back(path);
}
std::cout << "img_file_list length: " << img_files_list.size() << std::endl; std::cout << "img_file_list length: " << img_files_list.size() << std::endl;
Classifier classifier(config); Classifier classifier(config);
double elapsed_time = 0.0; double elapsed_time = 0.0;
std::vector<double> cls_times; std::vector<double> cls_times;
int warmup_iter = img_files_list.size() > 5 ? 5 : 0; int warmup_iter = img_files_list.size() > 5 ? 5 : 0;
for (int idx = 0; idx < img_files_list.size(); ++idx) { for (int idx = 0; idx < img_files_list.size(); ++idx) {
std::string img_path = img_files_list[idx]; std::string img_path = img_files_list[idx];
cv::Mat srcimg = cv::imread(img_path, cv::IMREAD_COLOR); cv::Mat srcimg = cv::imread(img_path, cv::IMREAD_COLOR);
if (!srcimg.data) { if (!srcimg.data) {
std::cerr << "[ERROR] image read failed! image path: " << img_path std::cerr << "[ERROR] image read failed! image path: " << img_path
<< "\n"; << "\n";
exit(-1); exit(-1);
} }
cv::cvtColor(srcimg, srcimg, cv::COLOR_BGR2RGB); cv::cvtColor(srcimg, srcimg, cv::COLOR_BGR2RGB);
double run_time = classifier.Run(srcimg, &cls_times); double run_time = classifier.Run(srcimg, &cls_times);
if (idx >= warmup_iter) { if (idx >= warmup_iter) {
elapsed_time += run_time; elapsed_time += run_time;
std::cout << "Current image path: " << img_path << std::endl; std::cout << "Current image path: " << img_path << std::endl;
std::cout << "Current time cost: " << run_time << " s, " std::cout << "Current time cost: " << run_time << " s, "
<< "average time cost in all: " << "average time cost in all: "
<< elapsed_time / (idx + 1 - warmup_iter) << " s." << std::endl; << elapsed_time / (idx + 1 - warmup_iter) << " s." << std::endl;
} else { } else {
std::cout << "Current time cost: " << run_time << " s." << std::endl; std::cout << "Current time cost: " << run_time << " s." << std::endl;
}
} }
}
std::string presion = "fp32"; std::string presion = "fp32";
if (config.use_fp16) if (config.use_fp16)
presion = "fp16"; presion = "fp16";
if (config.benchmark) { if (config.benchmark) {
AutoLogger autolog("Classification", config.use_gpu, config.use_tensorrt, AutoLogger autolog("Classification", config.use_gpu, config.use_tensorrt,
config.use_mkldnn, config.cpu_threads, 1, config.use_mkldnn, config.cpu_threads, 1,
"1, 3, 224, 224", presion, cls_times, "1, 3, 224, 224", presion, cls_times,
img_files_list.size()); img_files_list.size());
autolog.report(); autolog.report();
} }
return 0; return 0;
} }
...@@ -32,57 +32,57 @@ ...@@ -32,57 +32,57 @@
namespace PaddleClas { namespace PaddleClas {
void Permute::Run(const cv::Mat *im, float *data) { void Permute::Run(const cv::Mat *im, float *data) {
int rh = im->rows; int rh = im->rows;
int rw = im->cols; int rw = im->cols;
int rc = im->channels(); int rc = im->channels();
for (int i = 0; i < rc; ++i) { for (int i = 0; i < rc; ++i) {
cv::extractChannel(*im, cv::Mat(rh, rw, CV_32FC1, data + i * rh * rw), i); cv::extractChannel(*im, cv::Mat(rh, rw, CV_32FC1, data + i * rh * rw), i);
} }
} }
void Normalize::Run(cv::Mat *im, const std::vector<float> &mean, void Normalize::Run(cv::Mat *im, const std::vector<float> &mean,
const std::vector<float> &std, float &scale) { const std::vector<float> &std, float &scale) {
if (scale) { if (scale) {
(*im).convertTo(*im, CV_32FC3, scale); (*im).convertTo(*im, CV_32FC3, scale);
} }
for (int h = 0; h < im->rows; h++) { for (int h = 0; h < im->rows; h++) {
for (int w = 0; w < im->cols; w++) { for (int w = 0; w < im->cols; w++) {
im->at<cv::Vec3f>(h, w)[0] = im->at<cv::Vec3f>(h, w)[0] =
(im->at<cv::Vec3f>(h, w)[0] - mean[0]) / std[0]; (im->at<cv::Vec3f>(h, w)[0] - mean[0]) / std[0];
im->at<cv::Vec3f>(h, w)[1] = im->at<cv::Vec3f>(h, w)[1] =
(im->at<cv::Vec3f>(h, w)[1] - mean[1]) / std[1]; (im->at<cv::Vec3f>(h, w)[1] - mean[1]) / std[1];
im->at<cv::Vec3f>(h, w)[2] = im->at<cv::Vec3f>(h, w)[2] =
(im->at<cv::Vec3f>(h, w)[2] - mean[2]) / std[2]; (im->at<cv::Vec3f>(h, w)[2] - mean[2]) / std[2];
}
}
} }
}
}
void CenterCropImg::Run(cv::Mat &img, const int crop_size) { void CenterCropImg::Run(cv::Mat &img, const int crop_size) {
int resize_w = img.cols; int resize_w = img.cols;
int resize_h = img.rows; int resize_h = img.rows;
int w_start = int((resize_w - crop_size) / 2); int w_start = int((resize_w - crop_size) / 2);
int h_start = int((resize_h - crop_size) / 2); int h_start = int((resize_h - crop_size) / 2);
cv::Rect rect(w_start, h_start, crop_size, crop_size); cv::Rect rect(w_start, h_start, crop_size, crop_size);
img = img(rect); img = img(rect);
} }
void ResizeImg::Run(const cv::Mat &img, cv::Mat &resize_img, void ResizeImg::Run(const cv::Mat &img, cv::Mat &resize_img,
int resize_short_size) { int resize_short_size) {
int w = img.cols; int w = img.cols;
int h = img.rows; int h = img.rows;
float ratio = 1.f; float ratio = 1.f;
if (h < w) { if (h < w) {
ratio = float(resize_short_size) / float(h); ratio = float(resize_short_size) / float(h);
} else { } else {
ratio = float(resize_short_size) / float(w); ratio = float(resize_short_size) / float(w);
} }
int resize_h = round(float(h) * ratio); int resize_h = round(float(h) * ratio);
int resize_w = round(float(w) * ratio); int resize_w = round(float(w) * ratio);
cv::resize(img, resize_img, cv::Size(resize_w, resize_h)); cv::resize(img, resize_img, cv::Size(resize_w, resize_h));
} }
} // namespace PaddleClas } // namespace PaddleClas
...@@ -18,3 +18,4 @@ cmake .. \ ...@@ -18,3 +18,4 @@ cmake .. \
-DCUDA_LIB=${CUDA_LIB_DIR} \ -DCUDA_LIB=${CUDA_LIB_DIR} \
make -j make -j
cd ..
# model load config
use_gpu 0
gpu_id 0
gpu_mem 4000
cpu_threads 10
use_mkldnn 1
use_tensorrt 0
use_fp16 0
# cls config
cls_model_path /PaddleClas/inference/cls_infer.pdmodel
cls_params_path /PaddleClas/inference/cls_infer.pdiparams
resize_short_size 256
crop_size 224
# for log env info
benchmark 0
./build/clas_system ./tools/config.txt ./docs/imgs/ILSVRC2012_val_00000666.JPEG
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册