System.Runtime.InteropServices.SEHException (0x80004005): External component has thrown an exception. about facerecognitiondotnet HOT 4 CLOSED

@takuya-takeuchi
Could you show me minimal code to reproduce issue?

LossMetric_operator_matrixs is invoked when inference.
Therefore, I guess you try to invoke this methods from multiple thread.
If so, you must create multiple FaceRecognition object per thread.

Thanks.

from facerecognitiondotnet.

The code：
public static FaceRecognitionEx Inst
{
get
{
if (m_instance == null)
m_instance = new FaceRecognitionEx();
return m_instance;
}
}
private FaceRecognitionEx()
{
detector = Dlib.GetFrontalFaceDetector();
sp = ShapePredictor.Deserialize("shape_predictor_68_face_landmarks.dat");
net = DlibDotNet.Dnn.LossMetric.Deserialize("dlib_face_recognition_resnet_model_v1.dat");
}
FrontalFaceDetector detector;
ShapePredictor sp;
LossMetric net;
public List<float[]> GetFeature(string imgpath)
{
/*..........
var faces = new List<Matrix>();
try
{
using (var img = Dlib.LoadImageAsMatrix(imgpath))
{
foreach (var face in detector.Operator(img))
{
var shape = sp.Detect(img, face);
var faceChipDetail = Dlib.GetFaceChipDetails(shape, 150, 0.25);
var faceChip = Dlib.ExtractImageChip(img, faceChipDetail);

                        faces.Add(faceChip);
                    }
                    /*.......
            catch
            {
                LogManager.face.Info($"[获取人脸特征失败]");
                features = null;
            }
            foreach (var face in faces)
                face.Dispose();
        } while (false);
        return features;
  }

The function " GetFeature" was invoked by multiple thread.By my understanding,the “shape_predictor_68_face_landmarks.dat” loaded once can improve the performance.I have tried to load the .dat for each image,the cpu can't support.

from facerecognitiondotnet.

The function " GetFeature" was invoked by multiple thread.

In GetFeature, net.Operator is invoked from multiple thread, doesn't it? If so, this schenario is not be supported.
davisking/dlib#544 (comment)

You have to create FaceRecognitionEx instance for each thread.
Or you can invoke net.Operator on one thread by using lock mechanism.

DlibDotNet.Dnn.LossMetric would not be slow operation different from face detection.

from facerecognitiondotnet.

Thank you very much!net.Operator invoked from multiple thread lead to the problem.I have sovled the problem.

from facerecognitiondotnet.