Error using Marshal.Copy method

[omdxp]

Summary of your issue

Well, I was trying your example WebcamFacePose and then I got an error while using Marshal.Copy method that it said: "Argument 1 : cannot convert from 'System.Array' to 'int[]'"

Environment

.NET Framework 4.6.1

[takuya-takeuchi]

It is similar with #87?

[takuya-takeuchi]

@Omar-Belghaouti
Does it occur everytime?

If the 1st argument of Marshal.Copy is Int32[], 2nd argment must be Int32.
https://docs.microsoft.com/ja-jp/dotnet/api/system.runtime.interopservices.marshal.copy?view=netframework-4.6.1

But in https://github.com/takuya-takeuchi/DlibDotNet/blob/master/examples/WebcamFacePose/Program.cs#L52 , 1st is IntPtr and 2nd is byte[].
It should occur compile error before executing.

[omdxp]

No it's not throwing an exception, it's not running at all. When I use temp.Data as an argument it get me that error above.

The thing is I have a program written in python that use dlib library to detect hand with a trained model as below:

import dlib
import cv2

detector = dlib.simple_object_detector('hand.svm')

cv2.namedWindow('frame', cv2.WINDOW_NORMAL)

cap = cv2.VideoCapture(0)
rscale = 2.0

while (True):
    ret, frame = cap.read()
    width, height, _ = frame.shape

    ft = cv2.resize(frame, (int(frame.shape[1] / rscale), int(frame.shape[0] / rscale)))
    dets = detector(ft)

    for d in dets:
        cv2.rectangle(frame, (int(d.left()*rscale),
                              int(d.top()*rscale)),
                              (int(d.right()*rscale),
                              int(d.bottom()*rscale)),
                              (255, 0, 0), 5)

    cv2.imshow('frame', frame)

    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

cap.release()
cv2.destroyAllWindows()

And I want to do it with DlibDotNet

[omdxp]

Is this a way to avoid that error?

var temp = new Mat();
                    capture.Read(temp);

                    byte[] data = (byte[])temp.Data;
                    var array = new byte[temp.Width * temp.Height * temp.ElementSize];

                    int size = Marshal.SizeOf(data[0]) * data.Length;

                    IntPtr pnt = Marshal.AllocHGlobal(size);

                    try
                    {
                        Marshal.Copy(data, 0, pnt, data.Length);

                        Marshal.Copy(pnt, array, 0, array.Length);
                    }
                    finally
                    {
                        Marshal.FreeHGlobal(pnt);
                    }

[takuya-takeuchi]

byte[] data = (byte[])temp.Data;

Is it correct? It must be compile error.

[omdxp]

No it execute normally

[takuya-takeuchi]

Wait. What is Mat? I think Mat does not have ElementSize.

[omdxp]

I forgot to tell you that I'm using EmguCV wrapper not OpenCVSharp

[takuya-takeuchi]

Ops, I have never used EmguCV.
But EmguCV.Mat look like to have DataPointer property.
http://www.emgu.com/wiki/files/3.1.0/document/html/a4426997-e646-4f87-8ce7-793a6c62fc1b.htm

It may help you.

Marshal.Copy(temp.DataPointer, array, 0, array.Length);

[omdxp]

Thank you, I'll try it and tell you the results

[takuya-takeuchi]

Or
http://www.emgu.com/wiki/files/3.1.0/document/html/3fbad597-8949-3c0e-1477-a41a6f3a330c.htm

[omdxp]

I tried the first method temp.DataPointer and it works fine

[omdxp]

Could you add an example of hand detection using DlibDotNet the way as I did it in Python?

[omdxp]

Both methods works fine

[takuya-takeuchi]

I may be able to write sample code to detect hand.
But

• I do not know where is model file to detect human hand

Where did you get hand.svm file from? If license of model file is unknown, I can not accept your request?

[omdxp]

Here's the link :
https://github.com/TahaAnwar/opensource
It is the file "myhanddetector2.svm"

[takuya-takeuchi]

simple_object_detector is object_detector<scan_fhog_pyramid<pyramid_down<6>>>.
You can refere examples\FHogObjectDetector\Program.cs.

// Then you can recall it using the deserialize() function.
using (var tmp = new ScanFHogPyramid<PyramidDown, DefaultFHogFeatureExtractor>(6))
using (var detector2 = new ObjectDetector<ScanFHogPyramid<PyramidDown, DefaultFHogFeatureExtractor>>(tmp))
    detector2.Deserialize("face_detector.svm");

[omdxp]

Okay thanks, I'll try it. I really appreciate your help

[omdxp]

Here's what 's happening, when I used Deserialize method, the Operator method gives me this compile error:

The type arguments for method 'ObjectDetector<ScanFHogPyramid<PyramidDown, DefaultFHogFeatureExtractor>>.Operator<U>(Matrix<U>)' cannot be inferred from the usage. Try specifying the type arguments explicitly.

What should the type argument be?

[takuya-takeuchi]

ObjectDetector<ScanFHogPyramid<PyramidDown, DefaultFHogFeatureExtractor>>.Operator(Matrix matrix) accept Matrix.
U should be struct. eg; byte, int , RgbPixel.

detector2.Operator(new Matrix<byte>());

You can pass image data.

[omdxp]

It's done. I don't know if this is the perfect code for this but can you take look?

capture = new VideoCapture(0);
using (var win = new ImageWindow())
                {
                    using (var tmp = new ScanFHogPyramid<PyramidDown, DefaultFHogFeatureExtractor>(6))
                    using (var detector = new ObjectDetector<ScanFHogPyramid<PyramidDown, DefaultFHogFeatureExtractor>>(tmp))
                    {
                        detector.Deserialize("hand.svm");
                        while (!win.IsClosed())
                        {
                            var temp = new Mat();
                            capture.Read(temp);

                            var array = new byte[temp.Width * temp.Height * temp.ElementSize];
                            Marshal.Copy(temp.DataPointer, array, 0, array.Length);

                            using (var cimg = Dlib.LoadImageData<RgbPixel>(array, (uint)temp.Height, (uint)temp.Width, (uint)(temp.Width * temp.ElementSize)))
                            {
                                // Detect hands
                                var dets = detector.Operator(new DlibDotNet.Matrix<RgbPixel>(cimg));
                                
                                // Display it all on the screen
                                win.ClearOverlay();
                                win.SetImage(cimg);
                                win.AddOverlay(dets, new RgbPixel { Red = 255 });
                            }

                        }
                    }
                }

[omdxp]

But I noticed something, it is very slow comparing to my python program. Why is that?

[takuya-takeuchi]

var dets = detector.Operator(new DlibDotNet.Matrix<RgbPixel>(cimg));

you should use using statement for Matrix.

[takuya-takeuchi]

You did not use GPU?
DlibDotNet.Native does not use Intel MKL. Python dlib could use it.
So if use on cpu, DlibDotNet works on low performance.
Please refer #35.

You can build DlibDotNet.Native by using Intel MKL.

[omdxp]

Yes, I've used it on CPU, Thanks for your help.