iOS项目中OpenCV_contrib的Facemark模型文件获取方法咨询

Hey there! I see you're working on an iOS app using OpenCV and OpenCV_contrib for facial landmark detection, and you're stuck on getting the right model file for facemark->loadModel(). Let's break this down step by step.

1. Where to Get the FacemarkKazemi Model File #

The FacemarkKazemi algorithm requires a pre-trained model file to work correctly. This file is part of the OpenCV_contrib project's face module data:

• The standard filename is facemark_kazemi_model.dat (some older versions might use kazemi_landmark_model.dat or similar).
• You can find this file in the modules/face/data directory of the OpenCV_contrib repository—just locate and download that .dat file matching your OpenCV version.

2. Add the Model File to Your iOS Project #

Once you have the model file:

• Drag it directly into your Xcode project.
• Make sure to check "Copy items if needed" and select your app's target when prompted. This ensures the file is included in your app's bundle at runtime.

3. Update Your Code to Load the Model Correctly #

Right now, your code uses a placeholder string "load model" in facemark->loadModel()—we need to replace that with the actual path to the model file in your app bundle. Here's how to adjust your detectFacialLandmark: method:

(UIImage *) detectFacialLandmark:(UIImage *) image{
    double CurrentTime = CACurrentMediaTime();
    NSString *pathToFaceCascade = [[NSBundle mainBundle] pathForResource:@"lbpcascade_frontalface" ofType:@"xml"];
    // Add this line to get the landmark model path
    NSString *pathToLandmarkModel = [[NSBundle mainBundle] pathForResource:@"facemark_kazemi_model" ofType:@"dat"];
    
    cv::Mat faceImage;
    UIImageToMat(image, faceImage);
    cv::CascadeClassifier face_cascade;
    face_cascade.load([pathToFaceCascade UTF8String]);
    cv::face::FacemarkKazemi::Params params;
    cv::Ptr<cv::face::FacemarkKazemi> facemark = cv::face::FacemarkKazemi::create(params);
    facemark->setFaceDetector((cv::face::FN_FaceDetector)myDetector, &face_cascade);
    
    // Replace the placeholder with the actual model path
    facemark->loadModel([pathToLandmarkModel UTF8String]);
    
    std::vector<cv::Rect> faces;
    resize(faceImage,faceImage,cv::Size(460,460), 0, 0, cv::INTER_LINEAR_EXACT);
    facemark->getFaces(faceImage,faces);
    std::vector< std::vector<cv::Point2f> > shapes;
    if(facemark->fit(faceImage,faces,shapes))
    {
        for( size_t i = 0; i < faces.size(); i++ )
        {
            cv::rectangle(faceImage, faces[i], cvScalar(255, 0, 0));
        }
        for(unsigned long i=0;i<faces.size();i++){
            for(unsigned long k=0;k<shapes[i].size();k++)
                cv::circle(faceImage,shapes[i][k],5,cv::Scalar(0,0,255),cv::FILLED);
        }
    }
    UIImage *returnImage = MatToUIImage(faceImage);
    NSLog(@"test processing time : %f", (CACurrentMediaTime() - CurrentTime));
    return returnImage;
}

Important Notes #

• Version Matching: Ensure the model file matches your installed OpenCV/OpenCV_contrib version. Using a mismatched model can cause loading failures or inaccurate landmark detection.
• Path Debugging: If the model fails to load, print pathToLandmarkModel to the console—if it's nil, double-check that the file was added to your target and the filename in code matches the actual file name.
• Custom Compilation: If you compiled OpenCV/OpenCV_contrib from source, you can also include the model file in your build process to avoid manual addition.

内容的提问来源于stack exchange，提问作者Hsmlee