Hu Chunming / FFNvDecoder_WIN

#include "DogPoseDetectorOnnx.h"

#include "logger.h"
#include "ErrorRecorder.h"
#include "logging.h"
#include "common.h"

#include "cuda_kernels.h"
#include <algorithm>
#include <fstream>

#include "opencv2/opencv.hpp"

using namespace nvinfer1;

SampleErrorRecorder gRecorder;
namespace sample
{
	Logger gLogger{ Logger::Severity::kINFO };
	LogStreamConsumer gLogVerbose{ LOG_VERBOSE(gLogger) };
	LogStreamConsumer gLogInfo{ LOG_INFO(gLogger) };
	LogStreamConsumer gLogWarning{ LOG_WARN(gLogger) };
	LogStreamConsumer gLogError{ LOG_ERROR(gLogger) };
	LogStreamConsumer gLogFatal{ LOG_FATAL(gLogger) };

	void setReportableSeverity(Logger::Severity severity)
	{
		gLogger.setReportableSeverity(severity);
		gLogVerbose.setReportableSeverity(severity);
		gLogInfo.setReportableSeverity(severity);
		gLogWarning.setReportableSeverity(severity);
		gLogError.setReportableSeverity(severity);
		gLogFatal.setReportableSeverity(severity);
	}
} // namespace sample



DogPoseDetectorOnnx::DogPoseDetectorOnnx()
{
}


DogPoseDetectorOnnx::~DogPoseDetectorOnnx()
{
}

static size_t getFileSize(FILE* file) {
	size_t fileSize = -1;
	if (file != NULL) {
		if (fseek(file, 0L, SEEK_END) == 0) {
			fileSize = ftell(file);
		}
		rewind(file);
	}
	return fileSize;
}

bool DogPoseDetectorOnnx::init() {

	const char* model_path_onnx = "../weights/best.onnx";
	const char* input_node_name = "images";
	const char* output_node_name = "output";

	const char* plan_file_name = "dog_pose_detect_plan.bin";
	FILE* f = fopen(plan_file_name, "rb");
	if (f == nullptr)
	{
		auto builder = SampleUniquePtr<nvinfer1::IBuilder>(nvinfer1::createInferBuilder(sample::gLogger.getTRTLogger()));
		if (!builder)
		{
			return false;
		}

		const auto explicitBatch = 1U << static_cast<uint32_t>(NetworkDefinitionCreationFlag::kEXPLICIT_BATCH);
		auto network = SampleUniquePtr<nvinfer1::INetworkDefinition>(builder->createNetworkV2(explicitBatch));
		if (!network)
		{
			return false;
		}

		auto config = SampleUniquePtr<nvinfer1::IBuilderConfig>(builder->createBuilderConfig());
		if (!config)
		{
			return false;
		}

		auto parser = SampleUniquePtr<nvonnxparser::IParser>(nvonnxparser::createParser(*network, sample::gLogger.getTRTLogger()));
		if (!parser)
		{
			return false;
		}

		auto parsed = parser->parseFromFile(model_path_onnx, static_cast<int>(sample::gLogger.getReportableSeverity()));
		if (!parsed)
		{
			return false;
		}

		if (m_bUseFP16)
		{
			config->setFlag(BuilderFlag::kFP16);
		}
		else
		{
			config->setFlag(BuilderFlag::kINT8);
			samplesCommon::setAllDynamicRanges(network.get(), 127.0F, 127.0F);
		}

		samplesCommon::enableDLA(builder.get(), config.get(), m_dlaCore);


		// CUDA stream used for profiling by the builder.
		auto profileStream = samplesCommon::makeCudaStream();
		if (!profileStream)
		{
			return false;
		}
		config->setProfileStream(*profileStream);

		SampleUniquePtr<IHostMemory> plan{ builder->buildSerializedNetwork(*network, *config) };

		//plan = SampleUniquePtr<IHostMemory>(builder->buildSerializedNetwork(*network, *config));
		if (!plan)
		{
			return false;
		}

		FILE* fp = fopen(plan_file_name, "wb");
		fwrite(plan->data(), 1, plan->size(), fp);
		fclose(fp);

		mRuntime = std::shared_ptr<nvinfer1::IRuntime>(createInferRuntime(sample::gLogger.getTRTLogger()));
		if (!mRuntime)
		{
			return false;
		}

		mEngine = std::shared_ptr<nvinfer1::ICudaEngine>( mRuntime->deserializeCudaEngine(plan->data(), plan->size()), samplesCommon::InferDeleter());
		if (!mEngine)
		{
			return false;
		}
	}
	else {
		size_t file_size = getFileSize(f);
		void* pPlanData = malloc(file_size);

		fread(pPlanData, 1, file_size, f);

		bool bSucceed = false;
		do
		{
			mRuntime = std::shared_ptr<nvinfer1::IRuntime>(createInferRuntime(sample::gLogger.getTRTLogger()));
			if (!mRuntime)
			{
				break;
			}

			mEngine = std::shared_ptr<nvinfer1::ICudaEngine>( mRuntime->deserializeCudaEngine(pPlanData, file_size), samplesCommon::InferDeleter());
			if (!mEngine)
			{
				break;
			}

			bSucceed = true;
		} while (0);

		free(pPlanData);

		if (!bSucceed) {
			return false;
		}
	}

	context = SampleUniquePtr<nvinfer1::IExecutionContext>(mEngine->createExecutionContext());
	if (!context)
	{
		return false;
	}

	// 创建GPU显存缓冲区
	m_data_buffer = new void*[2];
	// 创建GPU显存输入缓冲区
	m_input_node_index = mEngine->getBindingIndex(input_node_name);
	m_input_node_dim = mEngine->getBindingDimensions(m_input_node_index);
	size_t input_data_length = m_input_node_dim.d[1] * m_input_node_dim.d[2] * m_input_node_dim.d[3];
	cudaMalloc(&(m_data_buffer[m_input_node_index]), input_data_length * sizeof(float));
	// 创建GPU显存输出缓冲区
	m_output_node_index = mEngine->getBindingIndex(output_node_name);
	m_output_node_dim = mEngine->getBindingDimensions(m_output_node_index);
	size_t output_data_length = m_output_node_dim.d[1] * m_output_node_dim.d[2];
	cudaMalloc(&(m_data_buffer[m_output_node_index]), output_data_length * sizeof(float));

	return true;
}

std::vector<DogPoseResult> DogPoseDetectorOnnx::detect(unsigned char *pGpuBgr, int src_width, int src_height) {
	int dst_width = m_input_node_dim.d[2];
	int dst_height = m_input_node_dim.d[3];

	int max_side_length = std::max(src_width, src_height);



	//saveCUDAImg(pGpuBgr, src_width, src_height, "src.jpg");

	cudaStream_t stream;
	cudaStreamCreate(&stream);

	{
		// 显存-->内存-->显存
		//int rgb_size = 3 * src_width * src_height;
		//uint8 *cpu_data = new uint8[rgb_size];
		//cudaError_t cudaStatus = cudaMemcpy(cpu_data, pGpuBgr, rgb_size * sizeof(uint8), cudaMemcpyDeviceToHost);
		//cv::Mat image(src_height, src_width, CV_8UC3, cpu_data);


		//cv::Mat max_image = cv::Mat::zeros(cv::Size(max_side_length, max_side_length), CV_8UC3);
		//cv::Rect roi(0, 0, image.cols, image.rows);
		//image.copyTo(max_image(roi));
		//// 将图像归一化，并放缩到指定大小
		//cv::Size input_node_shape(m_input_node_dim.d[2], m_input_node_dim.d[3]);
		//cv::Mat BN_image = cv::dnn::blobFromImage(max_image, 1 / 255.0, input_node_shape, cv::Scalar(0, 0, 0), true, false);

		//size_t input_data_length = m_input_node_dim.d[1] * m_input_node_dim.d[2] * m_input_node_dim.d[3];
		//std::vector<float> input_data(input_data_length);
		//memcpy(input_data.data(), BN_image.ptr<float>(), input_data_length * sizeof(float));

		//cudaMemcpyAsync(m_data_buffer[m_input_node_index], input_data.data(), input_data_length * sizeof(float), cudaMemcpyHostToDevice, stream);
	}

	cuda_common::resizeAndNorm(pGpuBgr, src_width, src_height, (float*)m_data_buffer[m_input_node_index], dst_width, dst_height);


	// 模型推理
	context->enqueueV2(m_data_buffer, stream, nullptr);

	size_t output_data_length = m_output_node_dim.d[1] * m_output_node_dim.d[2];
	float* result_array = new float[output_data_length];
	cudaMemcpyAsync(result_array, m_data_buffer[m_output_node_index], output_data_length * sizeof(float), cudaMemcpyDeviceToHost, stream);

	cudaDeviceSynchronize();


	ResultYolov5 result;
	result.factor = max_side_length / (float)m_input_node_dim.d[2];
	result.read_class_names("../weights/classes.txt");

	return result.yolov5_result(result_array, 0.6);
}