#ifndef _MODEL_H_
#define _MODEL_H_

#include "plugin_factory.h"
#include "trt_utils.h"
#include "calibrator.h"
#include "NvInfer.h"
#include "NvInferPlugin.h"
#include "NvInferRuntimeCommon.h"
#include "cuda_runtime_api.h"
#include <algorithm>
#include <cmath>
#include <stdint.h>
#include <string>
#include <vector>
#include "../utils/time_util.h"
#include "../config.h"
#include "opencv2/opencv.hpp"
#include <numeric>

struct NetworkInfo
{
    std::string networkType;
    std::string labelsFilePath;
    std::string precision;
    std::string deviceType;
    std::string calibrationTablePath;
    std::string enginePath;
    std::string inputBlobName;
    std::string data_path;
};

struct InferParams
{
    bool printPerfInfo;
    bool printPredictionInfo;
    std::string calibImages;
    std::string calibImagesPath;
    float probThresh;
    float nmsThresh;
};

struct TensorInfo
{
    std::string blobName;
    uint32_t stride{0};
    uint32_t stride_h{0};
    uint32_t stride_w{0};
    uint32_t gridSize{0};
    uint32_t grid_h{ 0 };
    uint32_t grid_w{ 0 };
    uint32_t numClasses{0};
    uint32_t numBBoxes{0};
    uint64_t volume{0};
    std::vector<uint32_t> masks;
    std::vector<float> anchors;
    int bindingIndex{-1};
    float* hostBuffer{nullptr};
};


class Detecter
{
public:
    float getNMSThresh() const { return m_NMSThresh; }
    std::string getClassName(const int& label) const { return m_ClassNames.at(label); }
    int getClassId(const int& label) const { return m_ClassIds.at(label); }
    uint32_t getInputH() const { return m_InputH; }
    uint32_t getInputW() const { return m_InputW; }
    uint32_t getNumClasses() const { return static_cast<uint32_t>(m_ClassNames.size()); }
    void doInference(const unsigned char* input, const uint32_t batchSize);
    std::vector<BBoxInfo> decodeDetections(const int& imageIdx,
                                            const int& imageH,
                                           const int& imageW);
    Detecter(const NetworkInfo& networkInfo, const InferParams& inferParams, int type);
    ~Detecter();

    std::string m_EnginePath;
    const std::string m_DeviceType;
    const std::string m_InputBlobName;
    std::vector<TensorInfo> m_OutputTensors;
    std::vector<std::map<std::string, std::string>> m_configBlocks;
    uint32_t m_InputH;
    uint32_t m_InputW;
    uint32_t m_InputC;
    uint64_t m_InputSize;
    uint32_t _n_classes = 0;
    float _f_depth_multiple = 0;
    float _f_width_multiple = 0;
    const float m_ProbThresh;
    const float m_NMSThresh;
    std::vector<std::string> m_ClassNames;
    const std::vector<int> m_ClassIds{
        1,  2,  3,  4,  5,  6,  7,  8,  9,  10, 11, 13, 14, 15, 16, 17, 18, 19, 20, 21,
        22, 23, 24, 25, 27, 28, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44,
        46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65,
        67, 70, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 84, 85, 86, 87, 88, 89, 90};
    uint32_t m_BatchSize = 1;
    nvinfer1::INetworkDefinition* m_Network;
    nvinfer1::IBuilder* m_Builder ;
    nvinfer1::IHostMemory* m_ModelStream;
    nvinfer1::ICudaEngine* m_Engine;
    nvinfer1::IExecutionContext* m_Context;
    std::vector<void*> m_DeviceBuffers;
    int m_InputBindingIndex;
    cudaStream_t m_CudaStream;
    PluginFactory* m_PluginFactory;

    std::vector<BBoxInfo> decodeTensor(const int imageIdx, const int imageH,
                                               const int imageW, const TensorInfo& tensor);

    BBox convert_bbox_res(const float& bx, const float& by, const float& bw, const float& bh,
        const uint32_t& stride_h_, const uint32_t& stride_w_, const uint32_t& netW, const uint32_t& netH)
    {
        BBox b;
        float x = bx * stride_w_;
        float y = by * stride_h_;

        b.x1 = x - bw / 2;
        b.x2 = x + bw / 2;

        b.y1 = y - bh / 2;
        b.y2 = y + bh / 2;

        b.x1 = clamp(b.x1, 0, netW);
        b.x2 = clamp(b.x2, 0, netW);
        b.y1 = clamp(b.y1, 0, netH);
        b.y2 = clamp(b.y2, 0, netH);

        return b;
    }


    inline void add_bbox_proposal(const float bx, const float by, const float bw, const float bh,
        const uint32_t stride_h_, const uint32_t stride_w_, const float scaleH, const float scaleW, const float xoffset_, const float yoffset, const int maxIndex, const float maxProb,
        const uint32_t  image_w, const uint32_t image_h,
        std::vector<BBoxInfo>& binfo)
    {
        BBoxInfo bbi;
        bbi.box = convert_bbox_res(bx, by, bw, bh, stride_h_, stride_w_, m_InputW, m_InputH);
        if ((bbi.box.x1 > bbi.box.x2) || (bbi.box.y1 > bbi.box.y2))
        {
            return;
        }
        bbi.box.x1 = ((float)bbi.box.x1 / (float)m_InputW)*(float)image_w;
        bbi.box.y1 = ((float)bbi.box.y1 / (float)m_InputH)*(float)image_h;
        bbi.box.x2 = ((float)bbi.box.x2 / (float)m_InputW)*(float)image_w;
        bbi.box.y2 = ((float)bbi.box.y2 / (float)m_InputH)*(float)image_h;

        bbi.label = maxIndex;
        bbi.prob = maxProb;
        bbi.classId = getClassId(maxIndex);
        binfo.push_back(bbi);
    };

    inline int64_t volume(const nvinfer1::Dims& d)
    {
        return std::accumulate(d.d,d.d+d.nbDims,1,std::multiplies<int64_t>());
    }
    inline unsigned int getElementSize(nvinfer1::DataType t)
    {
        switch (t)
        {
            case nvinfer1::DataType::kINT32: return 4;
            case nvinfer1::DataType::kFLOAT: return 4;
            case nvinfer1::DataType::kHALF: return 2;
            case nvinfer1::DataType::kBOOL:
            case nvinfer1::DataType::kINT8: return 1;
        }
        throw std::runtime_error("Invalid DataType.");
        return 0;
    }

    void setOutput(int type);
private:
    Logger m_Logger;
    void createYOLOEngine(const nvinfer1::DataType dataType = nvinfer1::DataType::kFLOAT,
                          Int8EntropyCalibrator* calibrator = nullptr);
    void writePlanFileToDisk();
    std::vector<std::map<std::string, std::string>> parseConfigFile(const std::string cfgFilePath);
    void parseConfigBlocks();
    void allocateBuffers();
    bool verifyEngine();
    void destroyNetworkUtils(std::vector<nvinfer1::Weights>& trtWeights);
protected:
    const std::string m_NetworkType;
    std::unique_ptr<YoloTinyMaxpoolPaddingFormula> m_TinyMaxpoolPaddingFormula;

private:
    Timer _timer;
    int _n_yolo_ind = 0;
};

#endif