Files
ANSCORE/modules/ANSODEngine/ANSONNXSEG.h
T
nghianguyenandClaude Opus 5 5820a5e61b feat(ODEngine): load a model from memory, not only from a path
Every engine built its session from a file path, which is why opening an
encrypted model package means inflating it to a directory first: the
unprotected model lands on disk, is optimised, and is zipped back into a
cache. The new package format cannot avoid that from the outside — the
engines have to be able to take bytes.

ANSModelSource.h introduces a ModelSource (a path, or bytes with a name
for logs) and one MakeSession that builds from either. The six ONNX
engines — classification, OBB, pose, segmentation, YOLO12 and YOLO —
each gain a load overload taking it, with the existing string form
delegating so no caller changes. The Windows wide-path conversion and
the from-memory call now live in one place instead of the eleven copies
they were spread across; YOLOOD, which had only the Windows branch,
becomes portable as a side effect.

The factories are named rather than implicit, so a call with a plain
string literal still resolves to the path overload without ambiguity.

Not compiled here: this layer builds only where CUDA is present, so
verification on this machine was the new header compiled and run, plus a
structural check that every engine declares and defines both overloads,
calls MakeSession exactly once, and has no raw session construction
left. It needs a build on a CUDA box before it ships.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-12 20:29:59 +10:00

90 lines
4.5 KiB
C++

#ifndef ANSONNXSEG_H
#define ANSONNXSEG_H
#pragma once
#include "ANSEngineCommon.h"
#include "ANSModelSource.h"
#include "engine.h"
namespace ANSCENTER {
class ANSENGINE_API ANSONNXSEG :public ANSODBase {
public:
virtual bool Initialize(std::string licenseKey, ModelConfig modelConfig, const std::string& modelZipFilePath, const std::string& modelZipPassword, std::string& labelMap) override;
virtual bool LoadModel(const std::string& modelZipFilePath, const std::string& modelZipPassword)override;
virtual bool LoadModelFromFolder(std::string licenseKey, ModelConfig modelConfig, std::string modelName, std::string className, const std::string& modelFolder, std::string& labelMap)override;
virtual bool OptimizeModel(bool fp16, std::string& optimizedModelFolder);
std::vector<Object> RunInference(const cv::Mat& input);
std::vector<Object> RunInference(const cv::Mat& input, const std::string& camera_id);
bool Destroy();
~ANSONNXSEG();
int getInstanceId() const { return instanceId_; }
private:
std::string _modelFilePath;
bool _modelLoadValid;
bool _fp16{ false };
size_t vectorProduct(const std::vector<int64_t>& vector);
void letterBox(const cv::Mat& image,
cv::Mat& outImage,
const cv::Size& newShape,
const cv::Scalar& color = cv::Scalar(114, 114, 114),
bool auto_ = true,
bool scaleFill = false,
bool scaleUp = true,
int stride = 32);
BoundingBox scaleCoords(const cv::Size& letterboxShape,
const BoundingBox& coords,
const cv::Size& originalShape,
bool p_Clip = true);
std::vector<cv::Scalar> generateColors(const std::vector<std::string>& classNames, int seed = 42);
cv::Mat sigmoid(const cv::Mat& src);
void NMSBoxes(const std::vector<BoundingBox>& boxes,
const std::vector<float>& scores,
float scoreThreshold,
float nmsThreshold,
std::vector<int>& indices);
void drawSegmentations(cv::Mat& image,
const std::vector<Object>& results,
float maskAlpha = 0.5f) const;
void drawSegmentationsAndBoxes(cv::Mat& image,
const std::vector<Object>& results,
float maskAlpha) const;
const std::vector<cv::Scalar>& getClassColors() const { return classColors; }
void warmupModel();
bool Init(const std::string& modelPath, bool useGPU=true, int deviceId = 0);
// Same load, from a model already in memory (an opened package).
bool Init(const ModelSource& model, bool useGPU = true, int deviceId = 0);
cv::Mat preprocess(const cv::Mat& image,float*& blobPtr,std::vector<int64_t>& inputTensorShape);
// outputs is non-const because GetTensorMutableData() requires a
// non-const Ort::Value receiver — see ANSONNXCL.h for full note.
std::vector<Object> postprocess(const cv::Size& origSize,const cv::Size& letterboxSize,
std::vector<Ort::Value>& outputs, const std::string& camera_id);
std::vector<Object> segment(const cv::Mat& image, const std::string& camera_id);
std::vector<cv::Point2f> maskToPolygon(const cv::Mat& binaryMask,
const cv::Rect& boundingBox,
float simplificationEpsilon = 2.0f,
int minContourArea = 10);
private:
static std::atomic<int> instanceCounter_; // Thread-safe counter
int instanceId_;
int deviceId_ = 0;
Ort::Env env{ nullptr }; // ONNX Runtime environment
Ort::SessionOptions sessionOptions{ nullptr }; // Session options for ONNX Runtime
Ort::Session session{ nullptr }; // ONNX Runtime session for running inference
bool isDynamicInputShape{}; // Flag indicating if input shape is dynamic
cv::Size inputImageShape; // Expected input image shape for the model
// Vectors to hold allocated input and output node names
std::vector<Ort::AllocatedStringPtr> inputNameAllocs;
std::vector<const char*> inputNames;
std::vector<Ort::AllocatedStringPtr> outputNameAllocs;
std::vector<const char*> outputNames;
size_t numInputNodes, numOutputNodes; // Number of input and output nodes in the model
std::vector<cv::Scalar> classColors;
float m_imgWidth = 0;
float m_imgHeight = 0;
};
}
#endif
#pragma once