ANSCORE/modules/ANSODEngine/ANSFaceDetectorEngine.cpp

#pragma once
#include "ANSODEngine.h"
#include "ANSYOLOOD.h"
#include "ANSTENSORRTOD.h"
#include "ANSTENSORRTCL.h"
#include "ANSOPENVINOCL.h"
#include "ANSOPENVINOOD.h"
#include "ANSYOLOV10RTOD.h"
#include "ANSYOLOV10OVOD.h"
#include "ANSCUSTOMDetector.h"
#include "ANSFD.h"
#include "ANSANOMALIB.h"
#include "ANSPOSE.h"
#include "ANSSAM.h"
#include <pipelines/metadata.h>
#include <models/input_data.h>
#include "utils/visualizer.hpp"
#include <turbojpeg.h>
#include <vector>
#include <map>
#include <string>
#define USEONNXOV
//#define ISVALIDFACE_DEBUG
namespace ANSCENTER
{
	void ANSFDBase::CheckLicense() {
		try {
			_licenseValid = ANSCENTER::ANSLicenseHelper::LicenseVerification(_licenseKey, 1004, "ANSFR");//Default productId=1004
		}
		catch (std::exception& e) {
			this->_logger.LogFatal("ANSFDBase::CheckLicense()", e.what(), __FILE__, __LINE__);
		}
	}
	ModelConfig  ANSFDBase::GetModelConfig() {
		return _modelConfig;
	}
	bool ANSFDBase::LoadModel(const std::string& modelZipFilePath, const std::string& modelZipPassword)
	{
		Cleanup();
		try {
			ANSCENTER::ANSLibsLoader::Initialize();
			_modelFolder = "";
			_modelConfigFile = "";
			_modelFolder.clear();
			_modelConfigFile.clear();
			// 0. Check if the modelZipFilePath exist?
			if (!FileExist(modelZipFilePath)) {
				this->_logger.LogFatal("ANSFDBase::LoadModel", "Model zip file is not exist", __FILE__, __LINE__);
			}

			// 1. Unzip model zip file to a special location with folder name as model file (and version)
			std::string outputFolder;
			std::vector<std::string> passwordArray;
			if (!modelZipPassword.empty()) passwordArray.push_back(modelZipPassword);
			passwordArray.push_back("Sh7O7nUe7vJ/417W0gWX+dSdfcP9hUqtf/fEqJGqxYL3PedvHubJag==");
			passwordArray.push_back("3LHxGrjQ7kKDJBD9MX86H96mtKLJaZcTYXrYRdQgW8BKGt7enZHYMg==");
			passwordArray.push_back("AnsCustomModels20@$");
			passwordArray.push_back("AnsDemoModels20@!");
			std::string modelName = GetFileNameWithoutExtension(modelZipFilePath);
			//this->_logger.LogDebug("ANSFDBase::LoadModel. Model name", modelName, __FILE__, __LINE__);
			size_t vectorSize = passwordArray.size();
			for (size_t i = 0; i < vectorSize; i++) {
				if (ExtractPasswordProtectedZip(modelZipFilePath, passwordArray[i], modelName, _modelFolder, false))
					break; // Break the loop when the condition is met. 
			}
			// 2. Check if the outputFolder exist
			if (!std::filesystem::exists(_modelFolder)) {
				this->_logger.LogError("ANSFDBase::LoadModel. Output model folder is not exist", _modelFolder, __FILE__, __LINE__);
				return false; // That means the model file is not exist or the password is not correct
			}
			// 3. Check if the model has the configuration file
			std::string modelConfigName = "model_config.json";
			_modelConfigFile = CreateFilePath(_modelFolder, modelConfigName);
			_facelivenessEngineValid = false;
			std::string livenessModelFile = "C:\\ProgramData\\ANSCENTER\\Shared\\ANSFaceAttributesModel.zip";
			if (FileExist(livenessModelFile)) {
				_facelivenessEngineValid = InitializeLivenessModel(_licenseKey, livenessModelFile, modelZipPassword);
			}
			if (_facelivenessEngineValid) {
				//Initialize face tracker
				CreateANSMOTHandle(&_faceTracker, 1);
				std::string params = "{\"parameters\":{\"frame_rate\":\"25\",\"track_buffer\" : \"60\",\"track_threshold\" : \"0.500000\",\"high_threshold\" : \"0.600000\",\"match_thresold\" : \"0.80000\"}}";
				UpdateANSMOTParameters(&_faceTracker, params.c_str());
#ifdef USE_TV_MODEL
				//Initialize tv tracker
				CreateANSMOTHandle(&_tvTracker, 1);
				std::string paramsTv = "{\"parameters\":{\"frame_rate\":\"25\",\"track_buffer\" : \"60\",\"track_threshold\" : \"0.500000\",\"high_threshold\" : \"0.600000\",\"match_thresold\" : \"0.80000\"}}";
				UpdateANSMOTParameters(&_tvTracker, paramsTv.c_str());
#endif
			}
			return true;  // Initialization successful
		}
		catch (std::exception& e) {
			this->_logger.LogFatal("ANSFDBase::LoadModel", e.what(), __FILE__, __LINE__);
			return false;
		}
	}
	bool ANSFDBase::OptimizeModel(bool fp16, std::string& optimizedModelFolder) {
		ANSCENTER::ANSLibsLoader::Initialize();
#ifdef USE_TV_MODEL
		if (_useTvDetector) {
			if (!this->_tvDetector->OptimizeModel(fp16, optimizedModelFolder)) return false;
			// Zip the optimized model folder
			std::string livenessModelFile = "C:\\ProgramData\\ANSCENTER\\Shared\\ANSFaceAttributesModel_Optimized.zip";
			if (FileExist(livenessModelFile)) return true;
			if (FolderExist(optimizedModelFolder)) {
				std::string modelZipPassword = "Sh7O7nUe7vJ/417W0gWX+dSdfcP9hUqtf/fEqJGqxYL3PedvHubJag==";
				return ZipFolderWithPassword(optimizedModelFolder.c_str(), livenessModelFile.c_str(), modelZipPassword.c_str());
			}
			return false;
		}
		else return true;  
#else
		return true;
#endif
	}
	bool ANSFDBase::Initialize(std::string licenseKey, ModelConfig modelConfig, const std::string& modelZipFilePath, const std::string& modelZipPassword, std::string& labelMap) {
		// Call Cleanup to ensure any existing resources are released before initialization
		Cleanup();
		try {
			ANSCENTER::ANSLibsLoader::Initialize();
			_licenseKey = licenseKey;
			_licenseValid = false;
			_modelFolder.clear();
			_modelConfigFile.clear();
			_imageProcessingModelFile.clear();

			//1. License check
			CheckLicense();
			if (!_licenseValid) {
				this->_logger.LogError("ANSFDBase::Initialize", "Invalid License", __FILE__, __LINE__);
				return false;
			}

			//2. Check if model zip file exists
			if (!FileExist(modelZipFilePath)) {
				this->_logger.LogFatal("ANSFDBase::Initialize", "Model zip file does not exist", __FILE__, __LINE__);
				return false;
			}

			//3. Unzip model file
			//_engineType = ANSLicenseHelper::CheckHardwareInformation();
			std::string modelName = GetFileNameWithoutExtension(modelZipFilePath);
			std::vector<std::string> passwordArray = { modelZipPassword,
													 "Sh7O7nUe7vJ/417W0gWX+dSdfcP9hUqtf/fEqJGqxYL3PedvHubJag==",
													 "3LHxGrjQ7kKDJBD9MX86H96mtKLJaZcTYXrYRdQgW8BKGt7enZHYMg==",
													 "AnsCustomModels20@$",
													 "AnsDemoModels20@!" };
			bool extractionSuccess = false;

			for (const auto& password : passwordArray) {
				if (ExtractPasswordProtectedZip(modelZipFilePath, password, modelName, _modelFolder, false)) {
					extractionSuccess = true;
					break;  // Break on successful extraction
				}
			}

			if (!extractionSuccess || !std::filesystem::exists(_modelFolder)) {
				this->_logger.LogError("ANSFDBase::Initialize. Output model folder does not exist", _modelFolder, __FILE__, __LINE__);
				return false;
			}
			_modelConfigFile = CreateFilePath(_modelFolder, "model_config.json");

			// 4. Load the image processing model
			//std::string imageProcessingModel = "C:\\ProgramData\\ANSCENTER\\Shared\\ANS_ImageProcessing_v1.0.zip";
			//if (FileExist(imageProcessingModel)) {
			//	std::string imagemodelName = GetFileNameWithoutExtension(imageProcessingModel);
			//	std::string _imageModelFolder;
			//	for (const auto& password : passwordArray) {
			//		if (ExtractPasswordProtectedZip(imageProcessingModel, password, imagemodelName, _imageModelFolder, false)) {
			//			extractionSuccess = true;
			//			break;  // Break on successful extraction
			//		}
			//	}
			//	_imageProcessingModelFile = CreateFilePath(_imageModelFolder, "upscale.xml");
			//	ov::Core core;
			//	if (FileExist(_imageProcessingModelFile)) {
			//		std::string deviceName = GetOpenVINODevice();
			//		try {
			//			if (_engineType == EngineType::NVIDIA_GPU) // NVIDIA CUDA
			//			{
			//				std::unique_ptr<ImageModel> model = std::unique_ptr<ImageModel>(new SuperResolutionModel(_imageProcessingModelFile, cv::Size(28, 28), ""));
			//				model->setInputsPreprocessing(false, "", "");
			//				_pipeline = new AsyncPipeline(std::move(model), ConfigFactory::getUserConfig(deviceName, 1, "", 4), core);

			//			}
			//			else {
			//				std::unique_ptr<ImageModel> model = std::unique_ptr<ImageModel>(new SuperResolutionModel(_imageProcessingModelFile, cv::Size(40, 40), ""));
			//				model->setInputsPreprocessing(false, "", "");
			//				_pipeline = new AsyncPipeline(std::move(model), ConfigFactory::getUserConfig(deviceName, 1, "", 4), core);
			//			}
			//		}
			//		catch (const std::exception& e) {
			//			// Fall back to CPU if GPU is not available
			//			if (_engineType == EngineType::NVIDIA_GPU) // NVIDIA CUDA
			//			{
			//				std::unique_ptr<ImageModel> model = std::unique_ptr<ImageModel>(new SuperResolutionModel(_imageProcessingModelFile, cv::Size(28, 28), ""));
			//				model->setInputsPreprocessing(false, "", "");
			//				_pipeline = new AsyncPipeline(std::move(model), ConfigFactory::getUserConfig("CPU", 1, "", 4), core);
			//			}
			//			else {
			//				std::unique_ptr<ImageModel> model = std::unique_ptr<ImageModel>(new SuperResolutionModel(_imageProcessingModelFile, cv::Size(40, 40), ""));
			//				model->setInputsPreprocessing(false, "", "");
			//				_pipeline = new AsyncPipeline(std::move(model), ConfigFactory::getUserConfig("CPU", 1, "", 4), core);
			//			}
			//		}		
			//	}

			//}
			// Initalize faceliveness model
			_facelivenessEngineValid = false;
			std::string livenessModelFile = "C:\\ProgramData\\ANSCENTER\\Shared\\ANSFaceAttributesModel_Optimized.zip";
			if (!FileExist(livenessModelFile)) livenessModelFile = "C:\\ProgramData\\ANSCENTER\\Shared\\ANSFaceAttributesModel.zip";

			if (FileExist(livenessModelFile)) {
				_facelivenessEngineValid = InitializeLivenessModel(_licenseKey, livenessModelFile, modelZipPassword);
			}
			if (_facelivenessEngineValid) {
				//Initialize face tracker
				CreateANSMOTHandle(&_faceTracker, 1);
				std::string params = "{\"parameters\":{\"frame_rate\":\"25\",\"track_buffer\" : \"60\",\"track_threshold\" : \"0.500000\",\"high_threshold\" : \"0.600000\",\"match_thresold\" : \"0.80000\"}}";
				UpdateANSMOTParameters(&_faceTracker, params.c_str());
#ifdef USE_TV_MODEL
				//Initialize tv tracker
				CreateANSMOTHandle(&_tvTracker, 1);
				std::string paramsTv = "{\"parameters\":{\"frame_rate\":\"25\",\"track_buffer\" : \"60\",\"track_threshold\" : \"0.500000\",\"high_threshold\" : \"0.600000\",\"match_thresold\" : \"0.80000\"}}";
				UpdateANSMOTParameters(&_tvTracker, paramsTv.c_str());
#endif
			}
			return true;  // Initialization successful
		}
		catch (const std::exception& e) {
			this->_logger.LogFatal("ANSFDBase::Initialize", e.what(), __FILE__, __LINE__);
			Cleanup();  // Ensure cleanup on failure
			return false;
		}
	}
	void ANSFDBase::Cleanup() {
		// Clear model paths and reset related data members
		_modelFolder.clear();
		_modelConfigFile.clear();
		_imageProcessingModelFile.clear();
		if (this->_faceTracker)
		{
			ReleaseANSMOTHandle(&_faceTracker);	
		}
		if (_livenessSession) {
			delete _livenessSession;
		}
		if (_ortLivenessSessionOptions) {
			delete _ortLivenessSessionOptions;
		}
		if (_ortLivenessEnv) {
			delete _ortLivenessEnv;
		}
#ifdef USE_TV_MODEL
		if (this->_tvTracker)
		{
			ReleaseANSMOTHandle(&_tvTracker);
		}
		if (_useTvDetector && this->_tvDetector) {
			this->_tvDetector->Destroy();
			this->_tvDetector.reset();
		}
#endif
	}
	bool ANSFDBase::LoadModelFromFolder(std::string licenseKey, ModelConfig modelConfig, std::string modelName, std::string className, const std::string& modelFolder, std::string& labelMap) {
		Cleanup();
		try {
			ANSCENTER::ANSLibsLoader::Initialize();
			_licenseKey = licenseKey;
			_licenseValid = false;
			_modelFolder = modelFolder;
			_modelConfigFile = "";
			_modelConfigFile.clear();
			// 0. Check license
			CheckLicense();
			if (!_licenseValid) {
				this->_logger.LogError("LoadModelFromFolder::Initialize", "Invalid License", __FILE__, __LINE__);
				return false;
			}
			// 1. Check if the modelFolder exist?
			if (!FolderExist(modelFolder)) {
				this->_logger.LogFatal("LoadModelFromFolder::Initialize", "Model zip file is not exist", __FILE__, __LINE__);
				return false;
			}
			// 3. Check if the model has the configuration file
			std::string modelConfigName = "model_config.json";
			_modelConfigFile = CreateFilePath(_modelFolder, modelConfigName);
			_facelivenessEngineValid = false;
			std::string livenessModelFile = "C:\\ProgramData\\ANSCENTER\\Shared\\ANSFaceAttributesModel.zip";
			if (FileExist(livenessModelFile)) {
				_facelivenessEngineValid = InitializeLivenessModel(_licenseKey, livenessModelFile, "");
			}
			if (_facelivenessEngineValid) {
				//Initialize face tracker
				CreateANSMOTHandle(&_faceTracker, 1);
				std::string params = "{\"parameters\":{\"frame_rate\":\"25\",\"track_buffer\" : \"60\",\"track_threshold\" : \"0.500000\",\"high_threshold\" : \"0.600000\",\"match_thresold\" : \"0.80000\"}}";
				UpdateANSMOTParameters(&_faceTracker, params.c_str());
#ifdef USE_TV_MODEL
				//Initialize tv tracker
				CreateANSMOTHandle(&_tvTracker, 1);
				std::string paramsTv = "{\"parameters\":{\"frame_rate\":\"25\",\"track_buffer\" : \"60\",\"track_threshold\" : \"0.500000\",\"high_threshold\" : \"0.600000\",\"match_thresold\" : \"0.80000\"}}";
				UpdateANSMOTParameters(&_tvTracker, paramsTv.c_str());
#endif
			}
			return true;  // Initialization successful
		}
		catch (std::exception& e) {
			this->_logger.LogFatal("LoadModelFromFolder::Initialize", e.what(), __FILE__, __LINE__);
			return false;
		}
		
	}
	cv::Mat ANSFDBase::Preprocess(cv::Mat& input_mat, std::vector<cv::Point2f>& face_landmark_5, cv::Mat& preprocessed_mat) {
		try {
			cv::Mat crop_image;
			cv::Mat affine_martix;
			std::tie(crop_image, affine_martix) = ANSUtilityHelper::AlignFacesSCRFD(input_mat, face_landmark_5);
			return crop_image;
		}
		catch (const std::exception& e) {
			this->_logger.LogFatal("ANSFDBase::Preprocess", e.what(), __FILE__, __LINE__);
			return cv::Mat();
		}
		catch (...) {
			this->_logger.LogFatal("ANSFDBase::Preprocess", "Unknown error occurred", __FILE__, __LINE__);
			return cv::Mat();
		}
	}
	std::string ANSFDBase::GetOpenVINODevice(ov::Core& core) {
		try {
			std::vector<std::string> available_devices = core.get_available_devices();
			bool device_found = false;
			std::string deviceName = "CPU";
			// Search for NPU
			auto it = std::find(available_devices.begin(), available_devices.end(), "NPU");
			if (it != available_devices.end()) {
				core.set_property("NPU", ov::hint::performance_mode(ov::hint::PerformanceMode::LATENCY));
				core.set_property("GPU", ov::hint::performance_mode(ov::hint::PerformanceMode::LATENCY));
				deviceName = "AUTO:NPU,GPU";
				device_found = true;
				return deviceName;
			}
			// If NPU not found, search for GPU
			if (!device_found) {
				it = std::find(available_devices.begin(), available_devices.end(), "GPU");
				if (it != available_devices.end()) {
					core.set_property("GPU", ov::hint::performance_mode(ov::hint::PerformanceMode::LATENCY));
					deviceName = "GPU";
					device_found = true;
					return deviceName;
				}
			}

			// If GPU not found, search for GPU.0
			if (!device_found) {
				it = std::find(available_devices.begin(), available_devices.end(), "GPU.0");
				if (it != available_devices.end()) {
					core.set_property("GPU", ov::hint::performance_mode(ov::hint::PerformanceMode::LATENCY));
					deviceName = "GPU";
					device_found = true;
					return deviceName;
				}
			}

			// If neither NPU nor GPU found, default to CPU
			if (!device_found) {
				core.set_property("CPU", ov::hint::performance_mode(ov::hint::PerformanceMode::LATENCY));
				deviceName = "CPU";
				return deviceName;
			}
			return deviceName;
		}
		catch (std::exception& e) {
			this->_logger.LogFatal("ANSFDBase::GetOpenVINODevice()", e.what(), __FILE__, __LINE__);
			core.set_property("CPU", ov::hint::performance_mode(ov::hint::PerformanceMode::LATENCY));
			return "CPU";
		}
	}
	cv::Mat ANSFDBase::GetCroppedFaceScale(const cv::Mat& image,
		const int x1, const int y1, const int x2, const int y2,
		int cropedImageSize) {
		std::lock_guard<std::recursive_mutex> lock(_mutex);
		cv::Mat resizedImage;
		cv::Mat frame = image.clone();
		try {
			// Check if the input image is valid
			if (image.empty() || !image.data) return resizedImage;

			// Calculate width and height of the face region
			int width = x2 - x1;
			int height = y2 - y1;

			// Validate the coordinates and dimensions
			if (width <= 0 || height <= 0) return resizedImage;
			if (x1 < 0 || y1 < 0 || x2 <= x1 || y2 <= y1) return resizedImage;

			// Define the bounding rectangle for cropping
			cv::Rect facePos(cv::Point(x1, y1), cv::Point(x2, y2));

			// Ensure the rectangle has valid dimensions
			if (facePos.width <= 0 || facePos.height <= 0) return resizedImage;

			// Apply Gaussian blur to the cropped region for smoother output
			cv::Mat tempCrop = frame(facePos).clone(); // Use clone to avoid modifying the input image

			// Resize the cropped face to the required size
			cv::resize(tempCrop, resizedImage, cv::Size(cropedImageSize, cropedImageSize), 0, 0, cv::INTER_LANCZOS4);

			// Release temporary resources
			tempCrop.release();
		}
		catch (const std::exception& e) {
			std::cerr << "Error in GetCroppedFaceScale: " << e.what() << std::endl;
		}

		// Release the cloned frame
		frame.release();

		return resizedImage;
	}


	ANSFDBase::~ANSFDBase() {
		try {
			Cleanup();
			_licenseValid = false;
			_licenseKey.clear();
			if (!_cameras.empty()) // Check if _cameras has data
			{
				for (auto& [key, cameraData] : _cameras)
				{
					cameraData.clear(); // Clear each CameraData instance
				}
				_cameras.clear(); // Clear the map itself
			}
		}
		catch (std::exception& e) {
			std::cout << "ANSFDBase::~ANSFDBase()" << e.what();
		}
	}
	bool ANSFDBase::isSimilarObject(const Object& obj1, const Object& obj2) {
		try {
			// If both objects have valid trackIds, compare based on trackId
			if (obj1.trackId != 0 && obj2.trackId != 0) {
				return obj1.trackId == obj2.trackId;
			}
			// Compare based on classId and className
			if (obj1.classId != obj2.classId ||
				obj1.className != obj2.className ||
				obj1.cameraId != obj2.cameraId) {
				return false;
			}
			// Define thresholds for similarity
			const double positionThreshold = 50.0;
			const double confidenceThreshold = 0.2;
			const double sizeThreshold = 0.2;  // Allows a 20% size difference

			// Check if centers are within positionThreshold
			cv::Point2f center1 = (obj1.box.tl() + obj1.box.br()) * 0.5;
			cv::Point2f center2 = (obj2.box.tl() + obj2.box.br()) * 0.5;
			if (cv::norm(center1 - center2) >= positionThreshold) {
				return false;
			}
			// Check if bounding box sizes are similar within the size threshold
			double widthRatio = static_cast<double>(obj1.box.width) / obj2.box.width;
			double heightRatio = static_cast<double>(obj1.box.height) / obj2.box.height;
			if (std::abs(1.0 - widthRatio) > sizeThreshold || std::abs(1.0 - heightRatio) > sizeThreshold) {
				return false;
			}
			// Check if confidence scores are within an acceptable range
			if (std::abs(obj1.confidence - obj2.confidence) >= confidenceThreshold) {
				return false;
			}
			// If all checks pass, the objects are considered similar
			return true;
		}
		catch (std::exception& e) {
			this->_logger.LogFatal("ANSFDBase::isSimilarObject", e.what(), __FILE__, __LINE__);
			return false;
		}

	}
	bool ANSFDBase::isOverlayObject(const Object& obj1, const Object& obj2) {
		try {
			// If both objects have valid trackIds, compare based on trackId
			if (obj1.trackId != 0 && obj2.trackId != 0) {
				return obj1.trackId == obj2.trackId;
			}
			// Compare based on classId and className
			if (obj1.classId != obj2.classId ||
				obj1.className != obj2.className ||
				obj1.cameraId != obj2.cameraId) {
				return false;
			}

			// Define thresholds for similarity
			const double positionThreshold = 50.0;

			// Check if centers are within positionThreshold
			cv::Point2f center1 = (obj1.box.tl() + obj1.box.br()) * 0.5;
			cv::Point2f center2 = (obj2.box.tl() + obj2.box.br()) * 0.5;
			if (cv::norm(center1 - center2) >= positionThreshold) {
				return false;
			}

			// Check if bounding boxes overlap
			if ((obj1.box & obj2.box).area() == 0) {
				return false;  // No overlap between the bounding boxes
			}

			// If all checks pass, the objects are considered similar
			return true;
		}
		catch (std::exception& e) {
			this->_logger.LogFatal("ANSFDBase::isOverlayObject", e.what(), __FILE__, __LINE__);
			return false;
		}
	}
	void ANSFDBase::EnqueueDetection(const std::vector<Object>& detectedObjects, const std::string& cameraId) {
		std::lock_guard<std::recursive_mutex> lock(_mutex);
		CameraData& camera = GetCameraData(cameraId);// Retrieve camera data
		if (camera._detectionQueue.size() == QUEUE_SIZE) {
			camera._detectionQueue.pop_front();  // Remove the oldest element if the queue is full
		}
		camera._detectionQueue.push_back(detectedObjects);  // Add the new detection, even if empty
	}
	std::deque<std::vector<Object>>ANSFDBase::DequeueDetection(const std::string& cameraId) {
		std::lock_guard<std::recursive_mutex> lock(_mutex);
		CameraData& camera = GetCameraData(cameraId);// Retrieve camera data
		return camera._detectionQueue;
	}
	std::vector<Object> ANSFDBase::DetectMovement(const cv::Mat& input, const std::string& camera_id) {
		std::lock_guard<std::recursive_mutex> lock(_mutex);
		std::vector<Object> objects;
		objects.clear();
		try {
			if (!_licenseValid) return objects;
			if (input.empty() || !input.data || !input.u) {
				return objects;
			}
			if ((input.cols < 5) || (input.rows < 5)) return objects;
			cv::Mat frame = input.clone();
			objects = this->_handler.MovementDetect(camera_id, frame);
			frame.release();
			return objects;
		}
		catch (std::exception& e) {
			this->_logger.LogFatal("ANSFDBase::DetectMovement", e.what(), __FILE__, __LINE__);
			return objects;
		}
	}
	cv::Rect ANSFDBase::GenerateMinimumSquareBoundingBox(const std::vector<ANSCENTER::Object>& detectedObjects, int minSize) {
		std::lock_guard<std::recursive_mutex> lock(_mutex);
		try {
			// Ensure there are rectangles to process
			int adMinSize = minSize - 20;
			if (adMinSize < 0) adMinSize = 0;
			if (detectedObjects.empty()) return cv::Rect(0, 0, minSize, minSize);

			// Initialize min and max coordinates
			int minX = detectedObjects[0].box.x;
			int minY = detectedObjects[0].box.y;
			int maxX = detectedObjects[0].box.x + detectedObjects[0].box.width;
			int maxY = detectedObjects[0].box.y + detectedObjects[0].box.height;

			// Calculate bounding box that includes all rectangles
			for (const auto& rect : detectedObjects) {
				minX = std::min(minX, rect.box.x);
				minY = std::min(minY, rect.box.y);
				maxX = std::max(maxX, rect.box.x + rect.box.width);
				maxY = std::max(maxY, rect.box.y + rect.box.height);
			}

			// Calculate width and height of the bounding box
			int width = maxX - minX;
			int height = maxY - minY;

			// Determine the size of the square
			int squareSize = std::max({ width, height, adMinSize });

			// Center the square around the bounding box
			int centerX = minX + width / 2;
			int centerY = minY + height / 2;
			int squareX = centerX - squareSize / 2;
			int squareY = centerY - squareSize / 2;

			return cv::Rect(squareX - 10, squareY - 10, squareSize + 20, squareSize + 20); // add 10 pixels to the square
		}
		catch (std::exception& e) {
			this->_logger.LogFatal("ANSFDBase::GenerateMinimumSquareBoundingBox", e.what(), __FILE__, __LINE__);
			return cv::Rect(0, 0, minSize, minSize);
		}
	}
	//bool ANSFDBase::isValidFace(const std::vector<cv::Point2f>& landmarks, const cv::Rect& faceRect, float maxEyeAngle, int offsetX, int offsetY) {
	//	try {
	//		// SCRFD should return exactly five landmarks in this order:
	//		// [0] left eye, [1] right eye, [2] nose, [3] left mouth, [4] right mouth.

	//		int faceWidth = faceRect.width;
	//		int faceHeight = faceRect.height;
	//		if ((faceWidth <= 25) || (faceHeight <= 25)) return false;
	//		if (landmarks.size() != 5)
	//			return false;
	//		std::vector<cv::Point2f> validLandmark;
	//		// Ensure all landmarks lie within the face rectangle.
	//		for (int i = 0; i < 5; i++)
	//		{
	//			cv::Point2f landmarkPoint = cv::Point2f(landmarks[i].x + offsetX, landmarks[i].y + offsetY);
	//			if (!faceRect.contains(landmarkPoint))return false;
	//			validLandmark.push_back(landmarkPoint);
	//		}

	//		// Extract landmarks with offsets
	//		cv::Point2f leftEye = validLandmark[0];
	//		cv::Point2f rightEye = validLandmark[1];
	//		cv::Point2f nose = validLandmark[2];
	//		cv::Point2f leftMouth = validLandmark[3];
	//		cv::Point2f rightMouth = validLandmark[4];

	//		// Compute the angle of the line connecting the eyes.
	//		float dx = rightEye.x - leftEye.x;
	//		float dy = rightEye.y - leftEye.y;
	//		float eyeAngle = std::atan2(dy, dx) * 180.0f / CV_PI;
	//		if (std::fabs(eyeAngle) > maxEyeAngle)
	//			return false; // Face is too tilted

	//		// Compute the interocular distance.
	//		float eyeDistance = std::hypot(dx, dy);

	//		// Compute mouth width.
	//		float mouthWidth = cv::norm(rightMouth - leftMouth);

	//		// Check the ratio between mouth width and eye distance.
	//		float ratio = mouthWidth / eyeDistance;
	//		if (ratio < 0.6f || ratio > 1.6f)
	//			return false;

	//		// --- Additional Nose Check ---

	//		// Compute the midpoint between the eyes.
	//		cv::Point2f eyeMidpoint((leftEye.x + rightEye.x) / 2.0f, (leftEye.y + rightEye.y) / 2.0f);

	//		// Check that the nose is near the horizontal midpoint.
	//		float noseDeviation = std::fabs(nose.x - eyeMidpoint.x);
	//		if (noseDeviation > eyeDistance * 0.2f)
	//			return false;

	//		// Check that the nose is vertically between the eyes and mouth.
	//		float eyeAvgY = (leftEye.y + rightEye.y) / 2.0f;
	//		float mouthAvgY = (leftMouth.y + rightMouth.y) / 2.0f;
	//		if (!(nose.y > eyeAvgY && nose.y < mouthAvgY))
	//			return false;

	//		return true;
	//	}
	//	catch (std::exception& e) {
	//		this->_logger.LogFatal("ANSFDBase::isValidFace", e.what(), __FILE__, __LINE__);
	//		return false;
	//	}

	//}
	bool ANSFDBase::isValidFace(const std::vector<cv::Point2f>& landmarks, const cv::Rect& faceRect,
		float maxEyeAngle, int offsetX, int offsetY, const cv::Mat& frame, float minBlurScore)
	{
		std::string rejectReason;

		// Early validation - no exception handling needed for simple checks
		if (faceRect.width <= 25 || faceRect.height <= 25) {
			rejectReason = "too_small";
#ifdef ISVALIDFACE_DEBUG
			std::cout << "[isValidFace] REJECT: face too small (" << faceRect.width << "x" << faceRect.height << ")" << std::endl;
#endif
			return false;
		}
		if (landmarks.size() != 5) {
			rejectReason = "bad_landmarks";
#ifdef ISVALIDFACE_DEBUG
			std::cout << "[isValidFace] REJECT: landmark count = " << landmarks.size() << " (expected 5)" << std::endl;
#endif
			return false;
		}
		try {
			// SCRFD landmarks: [0] left eye, [1] right eye, [2] nose, [3] left mouth, [4] right mouth
			const float offsetXf = static_cast<float>(offsetX);
			const float offsetYf = static_cast<float>(offsetY);

			// Apply offsets and validate all landmarks are within face rectangle
			const cv::Point2f leftEye(landmarks[0].x + offsetXf, landmarks[0].y + offsetYf);
			const cv::Point2f rightEye(landmarks[1].x + offsetXf, landmarks[1].y + offsetYf);
			const cv::Point2f nose(landmarks[2].x + offsetXf, landmarks[2].y + offsetYf);
			const cv::Point2f leftMouth(landmarks[3].x + offsetXf, landmarks[3].y + offsetYf);
			const cv::Point2f rightMouth(landmarks[4].x + offsetXf, landmarks[4].y + offsetYf);

			// Check all landmarks are within face rectangle
			if (!faceRect.contains(leftEye) || !faceRect.contains(rightEye) ||
				!faceRect.contains(nose) || !faceRect.contains(leftMouth) ||
				!faceRect.contains(rightMouth)) {
				rejectReason = "landmarks_outside";
#ifdef ISVALIDFACE_DEBUG
				std::cout << "[isValidFace] REJECT: landmarks outside face rect" << std::endl;
#endif
				// Fall through to debug visualization below
				goto debug_and_return;
			}

			{
			// Compute eye angle - reject if face is too tilted (roll check)
			const float dx = rightEye.x - leftEye.x;
			const float dy = rightEye.y - leftEye.y;
			const float eyeAngle = std::atan2(dy, dx) * (180.0f / static_cast<float>(CV_PI));

			if (std::fabs(eyeAngle) > maxEyeAngle) {
				rejectReason = "roll=" + std::to_string(eyeAngle);
#ifdef ISVALIDFACE_DEBUG
				std::cout << "[isValidFace] REJECT: roll angle = " << eyeAngle << " (max " << maxEyeAngle << ")" << std::endl;
#endif
				goto debug_and_return;
			}

			// Compute interocular distance
			const float eyeDistance = std::hypot(dx, dy);
			if (eyeDistance < 1.0f) {
				rejectReason = "eye_dist<1";
#ifdef ISVALIDFACE_DEBUG
				std::cout << "[isValidFace] REJECT: interocular distance too small = " << eyeDistance << std::endl;
#endif
				goto debug_and_return;
			}

			// Check mouth-to-eye ratio
			const float mouthWidth = std::hypot(rightMouth.x - leftMouth.x, rightMouth.y - leftMouth.y);
			const float ratio = mouthWidth / eyeDistance;

			if (ratio < 0.6f || ratio > 1.6f) {
				rejectReason = "mouth_ratio=" + std::to_string(ratio);
#ifdef ISVALIDFACE_DEBUG
				std::cout << "[isValidFace] REJECT: mouth/eye ratio = " << ratio << " (valid 0.6-1.6)" << std::endl;
#endif
				goto debug_and_return;
			}

			// Yaw estimation: asymmetry ratio of nose position between eyes
			// More robust than absolute pixel deviation — scale-invariant
			const float noseToLeft = std::abs(nose.x - leftEye.x);
			const float noseToRight = std::abs(rightEye.x - nose.x);
			const float asymmetry = std::abs(noseToLeft - noseToRight) / eyeDistance;

			if (asymmetry > 0.5f) {
				rejectReason = "yaw=" + std::to_string(asymmetry);
#ifdef ISVALIDFACE_DEBUG
				std::cout << "[isValidFace] REJECT: yaw asymmetry = " << asymmetry << " (max 0.5)" << std::endl;
#endif
				goto debug_and_return;
			}

			// Pitch estimation: nose vertical position relative to eye-mouth span
			const float eyeAvgY = (leftEye.y + rightEye.y) * 0.5f;
			const float mouthAvgY = (leftMouth.y + rightMouth.y) * 0.5f;
			const float faceHeight = mouthAvgY - eyeAvgY;

			if (faceHeight < 1.0f) {
				rejectReason = "faceH<1";
#ifdef ISVALIDFACE_DEBUG
				std::cout << "[isValidFace] REJECT: degenerate face height = " << faceHeight << std::endl;
#endif
				goto debug_and_return;
			}

			const float noseRelative = (nose.y - eyeAvgY) / faceHeight;
			if (noseRelative < 0.15f || noseRelative > 0.75f) {
				rejectReason = "pitch=" + std::to_string(noseRelative);
#ifdef ISVALIDFACE_DEBUG
				std::cout << "[isValidFace] REJECT: pitch noseRelative = " << noseRelative << " (valid 0.15-0.75)" << std::endl;
#endif
				goto debug_and_return;
			}

			// Optional blur check — only when frame is provided
			if (!frame.empty()) {
				cv::Rect safeRect = faceRect & cv::Rect(0, 0, frame.cols, frame.rows);
				if (safeRect.width > 10 && safeRect.height > 10) {
					cv::Mat gray;
					cv::cvtColor(frame(safeRect), gray, cv::COLOR_BGR2GRAY);
					cv::Mat lap;
					cv::Laplacian(gray, lap, CV_64F);
					cv::Scalar mean, stddev;
					cv::meanStdDev(lap, mean, stddev);
					double blurScore = stddev.val[0] * stddev.val[0];
					if (blurScore < static_cast<double>(minBlurScore)) {
						rejectReason = "blur=" + std::to_string(blurScore);
#ifdef ISVALIDFACE_DEBUG
						std::cout << "[isValidFace] REJECT: blur score = " << blurScore << " (min " << minBlurScore << ")" << std::endl;
#endif
						goto debug_and_return;
					}
				}
			}

#ifdef ISVALIDFACE_DEBUG
			std::cout << "[isValidFace] PASS: roll=" << eyeAngle << " yaw=" << asymmetry
				<< " pitch=" << noseRelative << " mouthRatio=" << ratio << std::endl;
#endif
			}

			// Debug visualization
			debug_and_return:
#ifdef ISVALIDFACE_DEBUG
			if (!frame.empty()) {
				cv::Rect safeRect = faceRect & cv::Rect(0, 0, frame.cols, frame.rows);
				if (safeRect.width > 10 && safeRect.height > 10) {
					// Crop and scale up for visibility
					cv::Mat debugImg = frame(safeRect).clone();
					const int debugSize = 256;
					float scale = static_cast<float>(debugSize) / std::max(debugImg.cols, debugImg.rows);
					cv::resize(debugImg, debugImg, cv::Size(), scale, scale);

					// Draw landmarks (scaled relative to face crop)
					auto toLocal = [&](const cv::Point2f& pt) -> cv::Point {
						return cv::Point(
							static_cast<int>((pt.x - safeRect.x) * scale),
							static_cast<int>((pt.y - safeRect.y) * scale));
					};

					// Landmark colors: green=eye, blue=nose, magenta=mouth
					cv::circle(debugImg, toLocal(leftEye), 3, cv::Scalar(0, 255, 0), -1);   // left eye
					cv::circle(debugImg, toLocal(rightEye), 3, cv::Scalar(0, 255, 0), -1);  // right eye
					cv::circle(debugImg, toLocal(nose), 3, cv::Scalar(255, 0, 0), -1);      // nose
					cv::circle(debugImg, toLocal(leftMouth), 3, cv::Scalar(255, 0, 255), -1);  // left mouth
					cv::circle(debugImg, toLocal(rightMouth), 3, cv::Scalar(255, 0, 255), -1); // right mouth

					// Draw eye line
					cv::line(debugImg, toLocal(leftEye), toLocal(rightEye), cv::Scalar(0, 200, 0), 1);
					// Draw mouth line
					cv::line(debugImg, toLocal(leftMouth), toLocal(rightMouth), cv::Scalar(200, 0, 200), 1);

					// PASS = green border, FAIL = red border + reason
					bool passed = rejectReason.empty();
					cv::Scalar borderColor = passed ? cv::Scalar(0, 255, 0) : cv::Scalar(0, 0, 255);
					cv::rectangle(debugImg, cv::Rect(0, 0, debugImg.cols, debugImg.rows), borderColor, 3);

					// Draw result text
					std::string label = passed ? "PASS" : "FAIL: " + rejectReason;
					cv::putText(debugImg, label, cv::Point(5, 20),
						cv::FONT_HERSHEY_SIMPLEX, 0.5, borderColor, 1);

					cv::imshow("isValidFace Debug", debugImg);
					cv::waitKey(1);
				}
			}
#endif
			return rejectReason.empty();

		}
		catch (const std::exception& e) {
			this->_logger.LogFatal("ANSFDBase::isValidFace", e.what(), __FILE__, __LINE__);
			return false;
		}
	}

	cv::Rect ANSFDBase::computeCandidateROI(const cv::Rect& unionBox, int fixedWidth, int fixedHeight, int imageWidth, int imageHeight) {
		// Compute center of the union.
		float centerX = unionBox.x + unionBox.width / 2.0f;
		float centerY = unionBox.y + unionBox.height / 2.0f;

		// Compute the required half-size in each dimension to cover the unionBox.
		float reqHalfWidth = std::max(centerX - static_cast<float>(unionBox.x), static_cast<float>(unionBox.x + unionBox.width) - centerX);
		float reqHalfHeight = std::max(centerY - static_cast<float>(unionBox.y), static_cast<float>(unionBox.y + unionBox.height) - centerY);

		// Desired full dimensions: at least fixed size, but as much as needed to cover unionBox.
		int desiredWidth = static_cast<int>(std::ceil(2 * reqHalfWidth));
		int desiredHeight = static_cast<int>(std::ceil(2 * reqHalfHeight));

		// Allow expansion up to 200% of fixed dimensions.
		int minWidth = fixedWidth;
		int minHeight = fixedHeight;
		int maxWidth = fixedWidth * 2;
		int maxHeight = fixedHeight * 2;

		int candidateWidth = std::min(std::max(desiredWidth, minWidth), maxWidth);
		int candidateHeight = std::min(std::max(desiredHeight, minHeight), maxHeight);

		// Compute top-left so that ROI is centered at (centerX, centerY).
		int roiX = static_cast<int>(std::round(centerX - candidateWidth / 2.0f));
		int roiY = static_cast<int>(std::round(centerY - candidateHeight / 2.0f));

		// Clamp the ROI to image boundaries.
		if (roiX < 0) roiX = 0;
		if (roiY < 0) roiY = 0;
		if (roiX + candidateWidth > imageWidth) roiX = imageWidth - candidateWidth;
		if (roiY + candidateHeight > imageHeight) roiY = imageHeight - candidateHeight;

		return cv::Rect(roiX, roiY, candidateWidth, candidateHeight);
	}
	std::vector<cv::Rect>   ANSFDBase::GenerateFixedROIs(const std::vector<Object>& movementObjects, int fixedWidth, int fixedHeight, int imageWidth, int imageHeight) {
		try {
			// Early check: if the image is smaller than the fixed ROI size, return one ROI covering the full image.
			if (imageWidth < fixedWidth || imageHeight < fixedHeight) {
				return { cv::Rect(0, 0, imageWidth, imageHeight) };
			}
			int maxImageSize = std::max(imageWidth, imageHeight);
			if (maxImageSize <= 1920) {
				int maxFixedSize = std::max(fixedWidth, fixedHeight);
				std::vector<cv::Rect> fixedROIs;
				cv::Rect   singleFixedROI = GenerateMinimumSquareBoundingBox(movementObjects, maxFixedSize);
				singleFixedROI.x = std::max(singleFixedROI.x, 0);
				singleFixedROI.y = std::max(singleFixedROI.y, 0);
				singleFixedROI.width = std::min(singleFixedROI.width, imageWidth - singleFixedROI.x);
				singleFixedROI.height = std::min(singleFixedROI.height, imageHeight - singleFixedROI.y);
				fixedROIs.push_back(singleFixedROI);
				return fixedROIs;
			}
			// --- Step 1: Group objects by proximity.
			// Use a threshold (fixedWidth/2) so that objects whose centers are within that distance belong to the same group.
			float groupingThreshold = fixedWidth / 2.0f;
			std::vector<Group> groups;

			for (const auto& obj : movementObjects) {
				cv::Point2f center(obj.box.x + obj.box.width / 2.0f, obj.box.y + obj.box.height / 2.0f);
				bool added = false;
				for (auto& grp : groups) {
					// Use the current group's center (from its unionBox).
					cv::Point2f groupCenter(grp.unionBox.x + grp.unionBox.width / 2.0f,
						grp.unionBox.y + grp.unionBox.height / 2.0f);
					if (distance(center, groupCenter) < groupingThreshold) {
						grp.objects.push_back(obj);
						grp.unionBox = unionRect(grp.unionBox, obj.box);
						added = true;
						break;
					}
				}
				if (!added) {
					Group newGroup;
					newGroup.objects.push_back(obj);
					newGroup.unionBox = obj.box;
					groups.push_back(newGroup);
				}
			}

			// --- Step 2: For each group, compute a candidate ROI that may expand up to 50% larger than fixed size.
			// Then merge groups whose candidate ROIs overlap.
			bool merged = true;
			while (merged) {
				merged = false;
				for (size_t i = 0; i < groups.size(); ++i) {
					cv::Rect roi_i = computeCandidateROI(groups[i].unionBox, fixedWidth, fixedHeight, imageWidth, imageHeight);
					for (size_t j = i + 1; j < groups.size(); ++j) {
						cv::Rect roi_j = computeCandidateROI(groups[j].unionBox, fixedWidth, fixedHeight, imageWidth, imageHeight);
						if (isOverlap(roi_i, roi_j)) {
							// Merge group j into group i.
							for (const auto& obj : groups[j].objects) {
								groups[i].objects.push_back(obj);
							}
							groups[i].unionBox = unionRect(groups[i].unionBox, groups[j].unionBox);
							groups.erase(groups.begin() + j);
							merged = true;
							break; // Break inner loop to restart merging.
						}
					}
					if (merged) break;
				}
			}

			// --- Step 3: Compute final candidate ROIs for each (merged) group.
			std::vector<cv::Rect> fixedROIs;
			for (const auto& grp : groups) {
				cv::Rect candidate = computeCandidateROI(grp.unionBox, fixedWidth, fixedHeight, imageWidth, imageHeight);
				fixedROIs.push_back(candidate);
			}

			// (Optional) sort the final ROIs.
			std::sort(fixedROIs.begin(), fixedROIs.end(), [](const cv::Rect& a, const cv::Rect& b) {
				return (a.y == b.y) ? (a.x < b.x) : (a.y < b.y);
				});

			return fixedROIs;
		}
		catch (std::exception& e) {
			this->_logger.LogFatal("ANSFDBase::GenerateFixedROIs", e.what(), __FILE__, __LINE__);
			return { cv::Rect(0, 0, imageWidth, imageHeight) };
		}

	}
	bool   ANSFDBase::ContainsIntersectingObject(const std::vector<Object>& movementObjects, const Object& result) {
		for (const auto& obj : movementObjects) {
			// Compute the intersection of the two bounding boxes.
			cv::Rect intersection = obj.box & result.box;
			if (intersection.area() > 0) {
				return true; // Found an intersecting object.
			}
		}
		return false; // No intersections found.
	}
	void   ANSFDBase::UpdateAndFilterDetectionObjects(std::vector<Object>& detectionObjects, int threshold) {
		detectionObjects.erase(
			std::remove_if(detectionObjects.begin(), detectionObjects.end(),
				[&](Object& obj) {
					if (!obj.extraInfo.empty()) {
						try {
							int value = std::stoi(obj.extraInfo); // Convert extraInfo to an integer
							if (value >= threshold) {
								return true; // Remove object if value exceeds threshold
							}
							obj.extraInfo = std::to_string(value + 1); // Increment extraInfo
						}
						catch (const std::exception&) {
							obj.extraInfo = "1"; // Reset to "1" if conversion fails
						}
					}
					return false; // Keep object
				}),
			detectionObjects.end()); // Remove flagged objects
	}

	bool    ANSFDBase::UpdateDetectionThreshold(float detectionScore) {
		try {
			this->_modelConfig.detectionScoreThreshold = detectionScore;
			return true;
		}
		catch (std::exception& e) {
			this->_logger.LogFatal("ANSFDBase::UpdateDetectionThreshold", e.what(), __FILE__, __LINE__);
			return false;
		}
	}
	float 	ANSFDBase::GetDetectionThreshold() {

		try {
			return this->_modelConfig.detectionScoreThreshold;
		}
		catch (std::exception& e) {
			this->_logger.LogFatal("ANSFDBase::GetDetectionThreshold", e.what(), __FILE__, __LINE__);
			return 0.0f;
		}
	}
	// Functions for screen size division
	double ANSFDBase::calculateDistanceToCenter(const cv::Point& center, const cv::Rect& rect) {
		cv::Point rectCenter(rect.x + rect.width / 2, rect.y + rect.height / 2);
		return std::sqrt(std::pow(rectCenter.x - center.x, 2) + std::pow(rectCenter.y - center.y, 2));
	}
	std::vector<ANSFDBase::ImageSection> ANSFDBase::divideImage(const cv::Mat& image) {
		if (image.empty()) {
			std::cerr << "Error: Empty image!" << std::endl;
			return cachedSections;
		}
		cv::Size currentSize(image.cols, image.rows);

		// Check if the image size has changed
		if (currentSize == previousImageSize) {
			return cachedSections; // Return cached sections if size is the same
		}

		// Update previous size
		previousImageSize = currentSize;
		cachedSections.clear();

		int width = image.cols;
		int height = image.rows;
		int maxDimension = std::max(width, height);
		int numSections = 10;// std::max(1, numSections); // Ensure at least 1 section
		if (maxDimension <= 2560)numSections = 8;
		if (maxDimension <= 1280)numSections = 6;
		if (maxDimension <= 960)numSections = 4;
		if (maxDimension <= 640)numSections = 2;
		if (maxDimension <= 320)numSections = 1;

		int gridRows = std::sqrt(numSections);
		int gridCols = (numSections + gridRows - 1) / gridRows; // Ensure all sections are covered

		int sectionWidth = width / gridCols;
		int sectionHeight = height / gridRows;

		cv::Point imageCenter(width / 2, height / 2);
		std::vector<std::pair<double, ImageSection>> distancePriorityList;

		// Create sections and store their distance from the center
		for (int r = 0; r < gridRows; ++r) {
			for (int c = 0; c < gridCols; ++c) {
				int x = c * sectionWidth;
				int y = r * sectionHeight;
				int w = (c == gridCols - 1) ? width - x : sectionWidth;
				int h = (r == gridRows - 1) ? height - y : sectionHeight;

				ImageSection section(cv::Rect(x, y, w, h));
				double distance = calculateDistanceToCenter(imageCenter, section.region);
				distancePriorityList.emplace_back(distance, section);
			}
		}

		// Sort sections based on distance from center, then top-to-bottom, then left-to-right
		std::sort(distancePriorityList.begin(), distancePriorityList.end(),
			[](const std::pair<double, ImageSection>& a, const std::pair<double, ImageSection>& b) {
				if (std::abs(a.first - b.first) > 1e-5) {
					return a.first < b.first; // Sort by closest distance to center
				}
				// If distance is the same, prioritize top to bottom, then left to right
				return a.second.region.y == b.second.region.y
					? a.second.region.x < b.second.region.x
					: a.second.region.y < b.second.region.y;
			});

		// Assign priority
		int priority = 1;
		for (auto& entry : distancePriorityList) {
			entry.second.priority = priority++;
			cachedSections.push_back(entry.second);
		}

		return cachedSections;
	}
	std::vector<ANSFDBase::ImageSection> ANSFDBase::createSlideScreens(const cv::Mat& image) {
		if (image.empty()) {
			std::cerr << "Error: Empty image!" << std::endl;
			return cachedSections;
		}

		cv::Size currentSize(image.cols, image.rows);
		if (currentSize == previousImageSize) {
			return cachedSections;
		}
		previousImageSize = currentSize;
		cachedSections.clear();

		int maxSize = std::max(image.cols, image.rows);
		const int minCellSize = 320;
		int maxSections = 10;
		const float minAspectRatio = 0.8f;
		const float maxAspectRatio = 1.2f;

		if (maxSize <= 640)	maxSections = 1;
		else if (maxSize <= 960) maxSections = 2;
		else if (maxSize <= 1280) maxSections = 4;
		else if (maxSize <= 2560) maxSections = 6;
		else if (maxSize <= 3840) maxSections = 8;
		else maxSections = 10;

		int width = image.cols;
		int height = image.rows;

		int bestRows = 1, bestCols = 1;
		int bestTileSize = std::numeric_limits<int>::max();

		for (int rows = 1; rows <= maxSections; ++rows) {
			for (int cols = 1; cols <= maxSections; ++cols) {
				if (rows * cols > maxSections) continue;

				int tileWidth = (width + cols - 1) / cols;
				int tileHeight = (height + rows - 1) / rows;

				if (tileWidth < minCellSize || tileHeight < minCellSize)
					continue;

				float aspectRatio = static_cast<float>(tileWidth) / static_cast<float>(tileHeight);
				if (aspectRatio < minAspectRatio || aspectRatio > maxAspectRatio)
					continue;

				int maxTileDim = std::max(tileWidth, tileHeight);
				if (maxTileDim < bestTileSize) {
					bestTileSize = maxTileDim;
					bestRows = rows;
					bestCols = cols;
				}
			}
		}

		// Generate tiles using bestRows, bestCols
		int tileWidth = (width + bestCols - 1) / bestCols;
		int tileHeight = (height + bestRows - 1) / bestRows;

		int priority = 1;
		for (int r = bestRows - 1; r >= 0; --r) {
			for (int c = 0; c < bestCols; ++c) {
				int x = c * tileWidth;
				int y = r * tileHeight;

				int w = std::min(tileWidth, width - x);
				int h = std::min(tileHeight, height - y);

				if (w <= 0 || h <= 0) continue;

				cv::Rect region(x, y, w, h);
				ImageSection section(region);
				section.priority = priority++;
				cachedSections.push_back(section);
			}
		}

		return cachedSections;
	}
	int ANSFDBase::getHighestPriorityRegion() {
		if (!cachedSections.empty()) {
			return cachedSections.front().priority; // First element has the highest priority
		}
		return 0; // Return empty rect if no sections exist
	}
	int ANSFDBase::getLowestPriorityRegion() {
		if (!cachedSections.empty()) {
			return cachedSections.back().priority; // Last element has the lowest priority
		}
		return 0; // Return empty rect if no sections exist
	}
	cv::Rect ANSFDBase::getRegionByPriority(int priority) {
		for (const auto& section : cachedSections) {
			if (section.priority == priority) {
				return section.region;
			}
		}
		return cv::Rect(); // Return empty rect if priority not found
	}

	std::vector<Object> ANSFDBase::AdjustDetectedBoundingBoxes(
		const std::vector<Object>& detectionsInROI,
		const cv::Rect& roi,
		const cv::Size& fullImageSize,
		float aspectRatio,
		int padding
	) {
		std::vector<Object> adjustedDetections;

		try {
			// Basic input validation
			if (detectionsInROI.empty()) {
				return adjustedDetections;
			}

			if (roi.width <= 0 || roi.height <= 0 ||
				fullImageSize.width <= 0 || fullImageSize.height <= 0) {
				return adjustedDetections;
			}

			for (const auto& detInROI : detectionsInROI) {
				try {
					if (detInROI.box.width <= 0 || detInROI.box.height <= 0)
						continue; // Skip invalid box

					// Convert ROI-relative box to full-image coordinates
					cv::Rect detInFullImg;
					try {
						detInFullImg = detInROI.box + roi.tl();
						detInFullImg &= cv::Rect(0, 0, fullImageSize.width, fullImageSize.height);
					}
					catch (const std::exception& e) {
						std::cerr << "[AdjustBBox] Failed to calculate full image box: " << e.what() << std::endl;
						continue;
					}

					// Check if it touches ROI border
					bool touchesLeft = detInROI.box.x <= 0;
					bool touchesRight = detInROI.box.x + detInROI.box.width >= roi.width - 1;
					bool touchesTop = detInROI.box.y <= 0;
					bool touchesBottom = detInROI.box.y + detInROI.box.height >= roi.height - 1;
					bool touchesBorder = touchesLeft || touchesRight || touchesTop || touchesBottom;

					// Compute target width based on aspect ratio
					int targetWidth = 0, expandWidth = 0;
					try {
						targetWidth = std::max(detInFullImg.width, static_cast<int>(detInFullImg.height * aspectRatio));
						expandWidth = std::max(0, targetWidth - detInFullImg.width);
					}
					catch (const std::exception& e) {
						std::cerr << "[AdjustBBox] Aspect ratio adjustment failed: " << e.what() << std::endl;
						continue;
					}

					int expandLeft = expandWidth / 2;
					int expandRight = expandWidth - expandLeft;
					int padX = touchesBorder ? padding : 0;
					int padY = touchesBorder ? (padding / 2) : 0;

					// Apply padded and expanded box
					int newX = std::max(0, detInFullImg.x - expandLeft - padX);
					int newY = std::max(0, detInFullImg.y - padY);
					int newWidth = detInFullImg.width + expandWidth + 2 * padX;
					int newHeight = detInFullImg.height + 2 * padY;

					// Clamp to image boundaries
					if (newX + newWidth > fullImageSize.width) {
						newWidth = fullImageSize.width - newX;
					}
					if (newY + newHeight > fullImageSize.height) {
						newHeight = fullImageSize.height - newY;
					}

					if (newWidth <= 0 || newHeight <= 0)
						continue;

					// Construct adjusted object
					Object adjustedDet = detInROI;
					adjustedDet.box = cv::Rect(newX, newY, newWidth, newHeight);
					adjustedDetections.push_back(adjustedDet);
				}
				catch (const std::exception& e) {
					std::cerr << "[AdjustBBox] Exception per detection: " << e.what() << std::endl;
					continue;
				}
				catch (...) {
					std::cerr << "[AdjustBBox] Unknown exception per detection." << std::endl;
					continue;
				}
			}
		}
		catch (const std::exception& e) {
			std::cerr << "[AdjustBBox] Fatal error: " << e.what() << std::endl;
		}
		catch (...) {
			std::cerr << "[AdjustBBox] Unknown fatal error occurred." << std::endl;
		}

		return adjustedDetections;
	}
	
	// Face detector engine implementation
	bool ANSFDBase::LoadLivenessModel(std::string antiSpoofModelPath, bool isGPU)
	{
		try {
			const auto& ep = ANSCENTER::EPLoader::Current();
			if (Ort::Global<void>::api_ == nullptr)
				Ort::InitApi(static_cast<const OrtApi*>(ANSCENTER::EPLoader::GetOrtApiRaw()));
			std::cout << "[ANSFDBase] EP ready: "<< ANSCENTER::EPLoader::EngineTypeName(ep.type) << std::endl;

			_ortLivenessEnv = new Ort::Env(ORT_LOGGING_LEVEL_WARNING, "Liveness");
			_ortLivenessSessionOptions = new Ort::SessionOptions();
			_ortLivenessSessionOptions->SetIntraOpNumThreads(
				std::min(6, static_cast<int>(std::thread::hardware_concurrency())));
			_ortLivenessSessionOptions->SetGraphOptimizationLevel(GraphOptimizationLevel::ORT_ENABLE_ALL);

			// ── Log available providers ─────────────────────────────────────────
			std::vector<std::string> availableProviders = Ort::GetAvailableProviders();
			std::cout << "[ANSFDBase] Available Execution Providers:" << std::endl;
			for (const auto& p : availableProviders)
				std::cout << " - " << p << std::endl;

			// ── Attach EP based on runtime-detected hardware ────────────────────
			if (isGPU) {
				bool attached = false;

				switch (ep.type) {

				case ANSCENTER::EngineType::NVIDIA_GPU: {
					auto it = std::find(availableProviders.begin(),
						availableProviders.end(), "CUDAExecutionProvider");
					if (it == availableProviders.end()) {
						this->_logger.LogError("ANSFDBase::LoadLivenessModel",
							"CUDAExecutionProvider not in DLL — "
							"check ep/cuda/ has the CUDA ORT build.", __FILE__, __LINE__);
						break;
					}
					try {
						OrtCUDAProviderOptionsV2* cuda_options = nullptr;
						Ort::GetApi().CreateCUDAProviderOptions(&cuda_options);

						const char* keys[] = { "device_id" };
						const char* values[] = { "0" };
						Ort::GetApi().UpdateCUDAProviderOptions(cuda_options, keys, values, 1);

						_ortLivenessSessionOptions->AppendExecutionProvider_CUDA_V2(*cuda_options);
						Ort::GetApi().ReleaseCUDAProviderOptions(cuda_options);

						std::cout << "[ANSFDBase] CUDA EP attached." << std::endl;
						attached = true;
					}
					catch (const Ort::Exception& e) {
						this->_logger.LogError("ANSFDBase::LoadLivenessModel", e.what(), __FILE__, __LINE__);
					}
					break;
				}

				case ANSCENTER::EngineType::AMD_GPU: {
					auto it = std::find(availableProviders.begin(),
						availableProviders.end(), "DmlExecutionProvider");
					if (it == availableProviders.end()) {
						this->_logger.LogError("ANSFDBase::LoadLivenessModel",
							"DmlExecutionProvider not in DLL — "
							"check ep/directml/ has the DirectML ORT build.", __FILE__, __LINE__);
						break;
					}
					try {
						std::unordered_map<std::string, std::string> opts = { { "device_id", "0" } };
						_ortLivenessSessionOptions->AppendExecutionProvider("DML", opts);
						std::cout << "[ANSFDBase] DirectML EP attached." << std::endl;
						attached = true;
					}
					catch (const Ort::Exception& e) {
						this->_logger.LogError("ANSFDBase::LoadLivenessModel", e.what(), __FILE__, __LINE__);
					}
					break;
				}

				case ANSCENTER::EngineType::OPENVINO_GPU: {
					auto it = std::find(availableProviders.begin(),
						availableProviders.end(), "OpenVINOExecutionProvider");
					if (it == availableProviders.end()) {
						this->_logger.LogError("ANSFDBase::LoadLivenessModel",
							"OpenVINOExecutionProvider not in DLL — "
							"check ep/openvino/ has the OpenVINO ORT build.", __FILE__, __LINE__);
						break;
					}

					const std::string precision = "FP16";
					const std::string numberOfThreads = "8";
					const std::string numberOfStreams = "8";

					std::vector<std::unordered_map<std::string, std::string>> try_configs = {
						{ {"device_type","AUTO:NPU,GPU"}, {"precision",precision},
						  {"num_of_threads",numberOfThreads}, {"num_streams",numberOfStreams},
						  {"enable_opencl_throttling","False"}, {"enable_qdq_optimizer","True"} },
						{ {"device_type","GPU.0"},        {"precision",precision},
						  {"num_of_threads",numberOfThreads}, {"num_streams",numberOfStreams},
						  {"enable_opencl_throttling","False"}, {"enable_qdq_optimizer","True"} },
						{ {"device_type","GPU.1"},        {"precision",precision},
						  {"num_of_threads",numberOfThreads}, {"num_streams",numberOfStreams},
						  {"enable_opencl_throttling","False"}, {"enable_qdq_optimizer","True"} },
						{ {"device_type","AUTO:GPU,CPU"}, {"precision",precision},
						  {"num_of_threads",numberOfThreads}, {"num_streams",numberOfStreams},
						  {"enable_opencl_throttling","False"}, {"enable_qdq_optimizer","True"} }
					};

					for (const auto& config : try_configs) {
						try {
							_ortLivenessSessionOptions->AppendExecutionProvider_OpenVINO_V2(config);
							std::cout << "[ANSFDBase] OpenVINO EP attached ("
								<< config.at("device_type") << ")." << std::endl;
							attached = true;
							break;
						}
						catch (const Ort::Exception& e) {
							this->_logger.LogError("ANSFDBase::LoadLivenessModel", e.what(), __FILE__, __LINE__);
						}
					}

					if (!attached)
						std::cerr << "[ANSFDBase] OpenVINO EP: all device configs failed." << std::endl;
					break;
				}

				default:
					break;
				}

				if (!attached) {
					std::cerr << "[ANSFDBase] No GPU EP attached — running on CPU." << std::endl;
					this->_logger.LogFatal("ANSFDBase::LoadLivenessModel",
						"GPU EP not attached. Running on CPU.", __FILE__, __LINE__);
				}
			}
			else {
				std::cout << "[ANSFDBase] Inference device: CPU (isGPU=false)" << std::endl;
			}

			// ── Load model ──────────────────────────────────────────────────────
			_livenessSession = new Ort::Session(
				*_ortLivenessEnv,
				std::wstring(antiSpoofModelPath.begin(), antiSpoofModelPath.end()).c_str(),
				*_ortLivenessSessionOptions);

			// ── Node names ──────────────────────────────────────────────────────
			Ort::AllocatorWithDefaultOptions allocator;

			Ort::AllocatedStringPtr input_name_ptr = _livenessSession->GetInputNameAllocated(0, allocator);
			_livenessInputName = input_name_ptr.get();

			Ort::AllocatedStringPtr output_name_ptr = _livenessSession->GetOutputNameAllocated(0, allocator);
			_livenessOutputName = output_name_ptr.get();

			std::cout << "[ANSFDBase] Liveness model loaded successfully." << std::endl;
			return true;
		}
		catch (const std::exception& e) {
			this->_logger.LogFatal("ANSFDBase::LoadLivenessModel", e.what(), __FILE__, __LINE__);
			return false;
		}
	}
	bool ANSFDBase::InitializeLivenessModel(std::string licenseKey, const std::string& modelZipFilePath, const std::string& modelZipPassword) {
		try {
			_faceAttrModelFolder.clear();

			if (!_licenseValid) {
				this->_logger.LogError("ANSFDBase::InitializeLivenessModel", "Invalid License", __FILE__, __LINE__);
				return false;
			}
			//2. Check if model zip file exists
			if (!FileExist(modelZipFilePath)) {
				this->_logger.LogFatal("ANSFDBase::InitializeLivenessModel", "Face Liveness model zip file does not exist", __FILE__, __LINE__);
				return false;
			}

			//3. Unzip model file
			//_engineType = ANSLicenseHelper::CheckHardwareInformation();
			std::string modelName = GetFileNameWithoutExtension(modelZipFilePath);
			std::vector<std::string> passwordArray = { modelZipPassword,
													 "Sh7O7nUe7vJ/417W0gWX+dSdfcP9hUqtf/fEqJGqxYL3PedvHubJag==",
													 "3LHxGrjQ7kKDJBD9MX86H96mtKLJaZcTYXrYRdQgW8BKGt7enZHYMg==",
													 "AnsCustomModels20@$",
													 "AnsDemoModels20@!" };
			bool extractionSuccess = false;

			for (const auto& password : passwordArray) {
				if (ExtractPasswordProtectedZip(modelZipFilePath, password, modelName, _faceAttrModelFolder, false)) {
					extractionSuccess = true;
					break;  // Break on successful extraction
				}
			}

			if (!extractionSuccess || !std::filesystem::exists(_faceAttrModelFolder)) {
				this->_logger.LogError("ANSFDBase::InitializeLivenessModel. Face Attribute model folder does not exist", _modelFolder, __FILE__, __LINE__);
				return false;
			}
			_useTvDetector = false;

#ifdef USE_TV_MODEL	
			// We can check to load TV model here (optional)
			ANSCENTER::ModelConfig _tvModelConfig;
			std::string tvLabelMap; 
			std::string tvModelPath = CreateFilePath(_faceAttrModelFolder, "screen.onnx");
			if (FileExist(tvModelPath)) {
				engineType = ANSCENTER::ANSLicenseHelper::CheckHardwareInformation();//  EngineType::CPU;// 
				if (engineType == ANSCENTER::EngineType::NVIDIA_GPU) {
					// Use TensorRT YoloV11
					_tvDetector = std::make_unique <ANSCENTER::ANSYOLOV10RTOD>();
				}
				else {
					// Use OpenVINO YoloV11
					_tvDetector = std::make_unique <ANSCENTER::ANSOYOLOV10OVOD>();
				}

				_tvModelConfig.detectionScoreThreshold = 0.6;
				_tvModelConfig.modelConfThreshold = 0.5;
				_tvModelConfig.modelMNSThreshold = 0.5;

				if (!_tvDetector->LoadModelFromFolder("", _tvModelConfig, "screen", "screen.names", _faceAttrModelFolder, tvLabelMap)) {
					this->_logger.LogError("ANSFDBase::InitializeLivenessModel", "Failed to load TV model from folder", __FILE__, __LINE__);
					_useTvDetector= false;
				}
				else _useTvDetector = true;
			}		
			// Perform tv model optimization if needed
			std::string optimizedModelFolder;
			if (_useTvDetector && (engineType == ANSCENTER::EngineType::NVIDIA_GPU)) {
				this->_tvDetector->OptimizeModel(true, optimizedModelFolder);
				// Zip the optimized model folder
				std::string livenessModelFile = "C:\\ProgramData\\ANSCENTER\\Shared\\ANSFaceAttributesModel_Optimized.zip";
				if (!FileExist(livenessModelFile)) {
					if (FolderExist(optimizedModelFolder)) {
						std::string modelZipPassword = "Sh7O7nUe7vJ/417W0gWX+dSdfcP9hUqtf/fEqJGqxYL3PedvHubJag==";
						ZipFolderWithPassword(optimizedModelFolder.c_str(), livenessModelFile.c_str(), modelZipPassword.c_str());
					}
				}
			}
#endif
			std::string antiSpoofModelPath = CreateFilePath(_faceAttrModelFolder, "liveness.onnx");
			if (!FileExist(antiSpoofModelPath)) {
				this->_logger.LogFatal("ANSFDBase::InitializeLivenessModel", "Liveness ONNX model file does not exist", __FILE__, __LINE__);
				return false;
			}
			return LoadLivenessModel(antiSpoofModelPath, true);
		}
		catch (const std::exception& e) {
			this->_logger.LogFatal("ANSFDBase::InitializeLivenessModel", e.what(), __FILE__, __LINE__);
			Cleanup();  // Ensure cleanup on failure
			return false;
		}
	}
	std::pair<int, float> ANSFDBase::PredictLiveness(const cv::Mat& image) {
		std::lock_guard<std::recursive_mutex> guard(_mutex);
		try {
			// 1. Resize
			cv::Mat faceFrame = image.clone();
			cv::resize(faceFrame, faceFrame, cv::Size(80, 80));

			// 2. Convert to float (keep BGR order and [0,255] range)
			faceFrame.convertTo(faceFrame, CV_32F, 1.0);

			// 3. Convert HWC to CHW
			std::vector<float> input_tensor_values(3 * 80 * 80);
			int idx = 0;
			for (int c = 0; c < 3; c++) {
				for (int y = 0; y < 80; y++) {
					for (int x = 0; x < 80; x++) {
						input_tensor_values[idx++] = faceFrame.at<cv::Vec3f>(y, x)[c];
					}
				}
			}

			std::vector<int64_t> input_shape = { 1, 3, 80, 80 };
			Ort::MemoryInfo memory_info = Ort::MemoryInfo::CreateCpu(OrtArenaAllocator, OrtMemTypeDefault);
			Ort::Value input_tensor = Ort::Value::CreateTensor<float>(
				memory_info, input_tensor_values.data(), input_tensor_values.size(),
				input_shape.data(), input_shape.size()
			);

			std::vector<const char*> input_names_arr{ _livenessInputName.c_str() };
			std::vector<const char*> output_names_arr{ _livenessOutputName.c_str() };

			auto output_tensors = _livenessSession->Run(
				Ort::RunOptions{ nullptr },
				input_names_arr.data(), &input_tensor, 1,
				output_names_arr.data(), 1
			);
			auto livenessResult = LivenessPostProcessing(output_tensors.front().GetTensorMutableData<float>());
			float bliveness = livenessResult.first;
			float conf = livenessResult.second;
			faceFrame.release();
			return std::pair<int, float>{bliveness, conf};
		}
		catch (std::exception& e) {
			this->_logger.LogFatal("ANSFDBase::PredictLiveness", e.what(), __FILE__, __LINE__);
			return std::pair<int, float>{ -1, 0.0f };
		}
	}
	std::pair<int, float> ANSFDBase::LivenessPostProcessing(const float* pOutput) {
		try {
			float max_val = *std::max_element(pOutput, pOutput + 3), sum = 0.0f, s[3];
			for (int i = 0; i < 3; ++i) sum += (s[i] = std::exp(pOutput[i] - max_val));
			for (int i = 0; i < 3; ++i) s[i] /= sum;

			auto max_it = std::max_element(s, s + 3);
			int label = static_cast<int>(std::distance(s, max_it));
			float confidence = *max_it;

			if (label != 1 && confidence > 0.6f) label = 0;
			return { label, confidence };
		}
		catch (std::exception& e) {
			this->_logger.LogFatal("ANSFDBase::LivenessPostProcessing", e.what(), __FILE__, __LINE__);
			return std::pair<int, float>{ -1, 0.0f };
		}
	}

	template<typename MapTrackData, typename MapMissingFrames>
	void ANSFDBase::CleanUpTracks(std::vector<Object>& currentObjects,
		MapTrackData& trackDataMap,
		MapMissingFrames& missingFramesMap,
		const int maxMissing,
		const int maxTracks)
	{
		std::set<int> currentIds;
		for (const auto& obj : currentObjects) {
			currentIds.insert(obj.trackId);
			missingFramesMap[obj.trackId] = 0;
		}
		// Increase missing counters
		for (auto& [trackId, count] : missingFramesMap) {
			if (currentIds.find(trackId) == currentIds.end()) {
				count++;
			}
		}
		// Remove stale tracks
		for (auto it = missingFramesMap.begin(); it != missingFramesMap.end();) {
			if (it->second >= maxMissing) {
				trackDataMap.erase(it->first);
				it = missingFramesMap.erase(it);
			}
			else {
				++it;
			}
		}
		// Enforce max track limit
		while (trackDataMap.size() > static_cast<size_t>(maxTracks)) {
			auto it = trackDataMap.begin(); // could remove lowest ID or use heuristic
			missingFramesMap.erase(it->first);
			trackDataMap.erase(it);
		}
	}
	
	float ANSFDBase::ComputeIoU(const cv::Rect& a, const cv::Rect& b) {
		int x1 = std::max(a.x, b.x);
		int y1 = std::max(a.y, b.y);
		int x2 = std::min(a.x + a.width, b.x + b.width);
		int y2 = std::min(a.y + a.height, b.y + b.height);

		int nInterWidth = std::max(0, x2 - x1);
		int nInterHeight = std::max(0, y2 - y1);
		int nInterArea = nInterWidth * nInterHeight;

		int nUnionArea = a.width * a.height + b.width * b.height - nInterArea;

		if (nUnionArea <= 0) return 0.0f;
		return static_cast<float>(nInterArea) / static_cast<float>(nUnionArea);
	}

	std::vector<Object>  ANSFDBase::TrackFaces(const cv::Mat& inputImage,const std::vector<Object> &faceObjects) {
		if (faceObjects.empty()) return faceObjects;
		if (!_facelivenessEngineValid) return faceObjects;
		std::vector<Object> vDetectedFaces;
		vDetectedFaces.reserve(faceObjects.size());
		std::lock_guard<std::recursive_mutex> guard(_mutex);
		try {
			std::vector<ANSCENTER::TrackerObject> vTrackerFace;
			//1. Prepare detected faces for tracking
			for (auto& obj : faceObjects) {
				vDetectedFaces.push_back(obj);
				// Update the tracker with the detected faces
				float pad_track_ratio = PAD_TRACK_RATIO;
				float pad_track_x = obj.box.width * pad_track_ratio;
				float pad_track_y = obj.box.height * pad_track_ratio;
				const int& class_id = obj.classId;
				const float& prob = obj.confidence;
				float x = std::max(0.0f, static_cast<float>(obj.box.x) - pad_track_x);
				float y = std::max(0.0f, static_cast<float>(obj.box.y) - pad_track_y);
				float width = std::min(obj.box.width + 2 * pad_track_x,
					static_cast<float>(inputImage.cols) - std::max(0.0f, obj.box.x - pad_track_x));
				float height = std::min(obj.box.height + 2 * pad_track_y,
					static_cast<float>(inputImage.rows) - std::max(0.0f, obj.box.y - pad_track_y));
				const float& left = x;
				const float& top = y;
				const float& right = x + width;
				const float& bottom = y + height;
				//// Create a ByteTrack::Object from the ANSCENTER::Object
				TrackerObject temp;
				temp.track_id = 0; // Placeholder, will be assigned by the tracker
				temp.prob = prob;
				temp.class_id = class_id;
				temp.x = left;
				temp.y = top;
				temp.width = width;
				temp.height = height;
				temp.left = left;
				temp.top = top;
				temp.right = right;
				temp.bottom = bottom;
				temp.object_id = obj.cameraId;
				vTrackerFace.push_back(temp);
			}
			//2. Update tracker and get tracked objects
			std::vector<TrackerObject> vObjTracker;
			UpdateANSMOTTracker(&_faceTracker, 0, vTrackerFace, vObjTracker);
			//3. Match tracker by	IoU with detected face
			for (const auto& trkr : vObjTracker) {
				int nTrackId = trkr.track_id;
				cv::Rect rTrkRect(
					static_cast<int>(trkr.x),
					static_cast<int>(trkr.y),
					static_cast<int>(trkr.width),
					static_cast<int>(trkr.height)
				);

				int nBestIdx = -1;
				float nBestIou = 0.0f;

				for (int i = 0; i < static_cast<int>(vDetectedFaces.size()); ++i) {
					float pad_track_ratio = PAD_TRACK_RATIO;
					int pad_x = static_cast<int>(vDetectedFaces[i].box.width * pad_track_ratio);
					int pad_y = static_cast<int>(vDetectedFaces[i].box.height * pad_track_ratio);

					cv::Rect paddedDetectBox(
						std::max(0, vDetectedFaces[i].box.x - pad_x),
						std::max(0, vDetectedFaces[i].box.y - pad_y),
						std::min(vDetectedFaces[i].box.width + 2 * pad_x, inputImage.cols - std::max(0, vDetectedFaces[i].box.x - pad_x)),
						std::min(vDetectedFaces[i].box.height + 2 * pad_y, inputImage.rows - std::max(0, vDetectedFaces[i].box.y - pad_y))
					);

					float fIou = ComputeIoU(rTrkRect, paddedDetectBox);
					if (fIou > nBestIou) {
						nBestIou = fIou;
						nBestIdx = i;
					}
				}
				// Only assign trackId if IoU is above threshold
				if (nBestIdx != -1 && nBestIou > 0.5f) {
					vDetectedFaces[nBestIdx].trackId = nTrackId;
				}
			}
			return vDetectedFaces;
		}
		catch (std::exception& e) {
			this->_logger.LogFatal("ANSFDBase::TrackFaces", e.what(), __FILE__, __LINE__);
			return vDetectedFaces;
		}
	}
#ifdef USE_TV_MODEL
	std::vector<Object>  ANSFDBase::TrackTVScreens(const std::vector<Object>& tvObjects) {
		if (tvObjects.empty()) return tvObjects;
		if (!_facelivenessEngineValid) return tvObjects;
		std::vector<Object> vRecallScreen;
		std::lock_guard<std::recursive_mutex> guard(_mutex);
		try {
			std::vector<Object> vDetectedScreens;
			vDetectedScreens.reserve(tvObjects.size());
			std::vector<ANSCENTER::TrackerObject> vTVTracker;
			//1. Prepare detected faces for tracking
			for (auto& obj : tvObjects) {
				vDetectedScreens.push_back(obj);
				// Update the tracker with the detected faces
				const int& class_id = obj.classId;
				const float& prob = obj.confidence;
				float x = obj.box.x;
				float y = obj.box.y;
				float width = obj.box.width;
				float height = obj.box.height;
				const float& left = x;
				const float& top = y;
				const float& right = x + width;
				const float& bottom = y + height;
				//// Create a ByteTrack::Object from the ANSCENTER::Object
				TrackerObject temp;
				temp.track_id = 0; // Placeholder, will be assigned by the tracker
				temp.prob = prob;
				temp.class_id = class_id;
				temp.x = left;
				temp.y = top;
				temp.width = width;
				temp.height = height;
				temp.left = left;
				temp.top = top;
				temp.right = right;
				temp.bottom = bottom;
				temp.object_id = obj.cameraId;
				vTVTracker.push_back(temp);
			}
			//2. Update tracker and get tracked objects
			std::vector<TrackerObject> vObjTracker;
			UpdateANSMOTTracker(&_tvTracker, 0, vTVTracker, vObjTracker);
			//3. Match tracker by	IoU with detected face
			for (const auto& trkr : vObjTracker) {
				int nTrackId = trkr.track_id;
				cv::Rect rTrkRect(
					static_cast<int>(trkr.x),
					static_cast<int>(trkr.y),
					static_cast<int>(trkr.width),
					static_cast<int>(trkr.height)
				);

				int nBestIdx = -1;
				float nBestIou = 0.0f;

				for (int i = 0; i < static_cast<int>(vDetectedScreens.size()); ++i) {
					cv::Rect DetectBox(
						vDetectedScreens[i].box.x,
						vDetectedScreens[i].box.y,
						vDetectedScreens[i].box.width,
						vDetectedScreens[i].box.height);

					float fIou = ComputeIoU(rTrkRect, DetectBox);
					if (fIou > nBestIou) {
						nBestIou = fIou;
						nBestIdx = i;
					}
				}
				// Only assign trackId if IoU is above threshold
				if (nBestIdx != -1 && nBestIou > 0.5f) {
					vDetectedScreens[nBestIdx].trackId = nTrackId;
				}
			}

			for (auto& screen : vDetectedScreens) {
				if (screen.trackId != -1) {
					if (screen.trackId < 0) continue;
					auto& ScreenHistory = _mTrackScreen[screen.trackId];
					ScreenHistory.push_back(screen);
				}
			}

			vRecallScreen.reserve(_mTrackScreen.size());
			for (auto& [id, objects] : _mTrackScreen) {
				if (!objects.empty()) {
					vRecallScreen.push_back(objects.back());
				}
			}
			CleanUpTracks(vDetectedScreens, _mTrackScreen, _mMissingTrackScreen, MAX_MISSING_SCREEN, 1);
			return vRecallScreen;
		}
		catch (std::exception& e) {
			this->_logger.LogFatal("ANSFDBase::TrackTVScreens", e.what(), __FILE__, __LINE__);
			return vRecallScreen;
		}
	}
	bool ANSFDBase::InsideScreen(const cv::Rect& a, const cv::Rect& b) {
		if (a.x <= b.x && a.y <= b.y && a.width >= b.width && a.height >= b.height)
			return true;
		return false;
	}
#endif

	std::vector<Object> ANSFDBase::ValidateLivenessFaces(const cv::Mat& inputImage,const std::vector<Object> &faceObjects, const std::string& camera_id) {
		if (!_facelivenessEngineValid) return faceObjects;
		if (faceObjects.empty()) return faceObjects;	
		std::lock_guard<std::recursive_mutex> guard(_mutex);
		try {
			std::vector<Object> results;
			// 1. Track faces
			std::vector<Object> vDetectedFaces = TrackFaces(inputImage, faceObjects);
			
			// 2. Run model liveness detection (vDetectedFaces with trackId assigned)
			for (auto& result : vDetectedFaces) {
				float fpad_ratio = PAD_DETECT_RATIO;
				int npad_x = static_cast<int>(result.box.width * fpad_ratio);
				int npad_y = static_cast<int>(result.box.height * fpad_ratio);

				cv::Rect paddedROI(
					std::max(0, result.box.x - npad_x),
					std::max(0, result.box.y - npad_y),
					std::min(result.box.width + 2 * npad_x, inputImage.cols - std::max(0, result.box.x - npad_x)),
					std::min(result.box.height + 2 * npad_y, inputImage.rows - std::max(0, result.box.y - npad_y))
				);
				// Crop the face with padding
				cv::Rect faceROI = paddedROI & cv::Rect(0, 0, inputImage.cols, inputImage.rows);
				if (faceROI.width <= 0 || faceROI.height <= 0) continue;

				cv::Mat face = inputImage(faceROI).clone();
				std::pair<int, float> prLiveness = PredictLiveness(face);
				//Return 1 if face is real, else return 0. Use for tracking attribute liveness
				int bliveness = prLiveness.first;
				
				//Skip the empty face input
				if (bliveness == -1) {
					continue;
				}

				//Verify faces on screens(TV, laptop, etc.)
				bool bisFake = false;
#ifdef USE_TV_MODEL
				if (_useTvDetector) {
					std::vector<Object> tvObjects= _tvDetector->RunInference(inputImage,camera_id); // We need to run inference here to get TV objects
					std::vector<Object> vRecallScreen = TrackTVScreens(tvObjects);
					for (const auto& obj : vRecallScreen) {
						if (InsideScreen(obj.box, result.box))
							bisFake = true;
					}
				}
#endif
				// If the face (with trackId) has bliveness for MAX_HISTORY_FACE frames, we can determine real/fake
				if (result.trackId != -1) {
					auto& dqHistory = _mTrackHistory[result.trackId];
					dqHistory.push_back(prLiveness.first);
					if (dqHistory.size() >= MAX_HISTORY_FACE) {
						int sumDeque = std::accumulate(dqHistory.begin(), dqHistory.end(), 0);
						// Decide "Real" if enough real attributes and face not on Screen else "Fake"
						if (sumDeque == MAX_HISTORY_FACE && !bisFake)
							result.extraInfo = "real";
						else
							result.extraInfo = "fake";
						results.push_back(result);
						dqHistory.pop_front();
					}
					else {
						result.extraInfo = "unknown";
						results.push_back(result);
					}
				}
			}
			CleanUpTracks(vDetectedFaces, _mTrackHistory, _mMissingTrackFrames, MAX_MISSING_FACE, MAX_TRACKS);

			return results;
		}
		catch (std::exception& e) {
			
			this->_logger.LogFatal("ANSFDBase::ValidateLivenessFaces", e.what(), __FILE__, __LINE__);
			return faceObjects;
		}
	}

}