Skip to content

README.en.md

Latest commit

 

History

History
130 lines (94 loc) · 3.94 KB

README.en.md

File metadata and controls

130 lines (94 loc) · 3.94 KB

DeepFaceSharp

en | 中文

DeepFaceSharp is a C#-based facial recognition library, which is a C# implementation of deepface.

The models in this project are derived from those used in deepface. To use them in .NET, they have been converted to ONNX format. Models can be automatically downloaded from GitHub.

OpenCVSharp is used for image processing.

Currently, the ported face detection algorithms include:

  • yunet
  • yolo8
  • yolo11

Face recognition algorithms include:

  • arcface
  • facenet

The project can directly use the specific algorithms mentioned above, or can be used based on the DeepFace class.

Usage Examples

Face Detection Algorithms

yunet

    var yunet = new YuNet();
    var fileName = "../../../../SampleImages/000000001000.jpg";

    var image = Cv2.ImRead(fileName);
    var ret = yunet.DetectFaces(image);
    Console.WriteLine("find face: " + ret.Count);
    foreach (var face in ret)
    {
        DrawFace(image, face);
    }

    Cv2.ImWrite(new FileInfo(fileName).Name.Replace(".jpg", "_YuNet.jpg"), image);

yolo8

    ModelConfiguration.Instance.SetDetectionThreshold(0.6f);
    var yunet = new Yolo();
    var fileName = "../../../../SampleImages/000000001000.jpg";

    var image = Cv2.ImRead(fileName);
    var ret = yunet.DetectFaces(image);
    Console.WriteLine(ret.Count);
    foreach (var face in ret)
    {
        DrawFace(image, face);
    }

    // Save the marked image
    Cv2.ImWrite(new FileInfo(fileName).Name.Replace(".jpg", "_yolo8.jpg"), image);

yolo11

  • Note: yolo11 algorithm's return does not include the 5-point facial data
    ModelDownloadUtils.ProxyServer = "http://192.168.1.3:10811";
    ModelConfiguration.Instance.SetDetectionThreshold(0.6f);
    var yunet = new Yolo(Yolo.YoloModel.YoloV11NFace);

    var fileName = "../../../../SampleImages/000000001000.jpg";
    var image = Cv2.ImRead(fileName);
    var ret = yunet.DetectFaces(image);
    Console.WriteLine(ret.Count);
    foreach (var face in ret)
    {
        DrawFace(image, face);
    }

    Cv2.ImWrite(new FileInfo(fileName).Name.Replace(".jpg", "_yolo11.jpg"), image);

Face Recognition Algorithms

arcface

    var arcFace = new ArcFace();
    var fileName = "../../../../SampleImages/000000001000_align_False_4.jpg";

    var image = Cv2.ImRead(fileName);
    var ret = arcFace.GetEmbeddingByMat(image);
    Console.WriteLine(ret.Length);
    var txt = string.Join("\n", ret);
    File.WriteAllText("arcface_net_onnx.txt", txt);

facenet

    var facenet = new Facenet();
    var fileName = "../../../../SampleImages/000000001000_align_False_4.jpg";

    var image = Cv2.ImRead(fileName);
    var ret = facenet.GetEmbeddingByMat(image);
    Console.WriteLine(ret.Length);
    var txt = string.Join("\n", ret);
    File.WriteAllText("facenet128_net_onnx.txt", txt);

GPU Version Adaptation

Although different versions of the ONNX CUDA runtime support from 11.8 to 12.x details, the .NET compiled versions do not have such broad version support. Starting from version 1.20.0, CUDA support is limited to 12.x and requires cudnn9 or above.

ONNX Runtime Version CUDA Version Notes
1.16.x - 1.19.x CUDA 11.8 Requires cudnn8
1.20.0 and above CUDA 12.x Requires cudnn9
  • If CUDA cannot be enabled, it's recommended to choose an appropriate ONNX version according to the version correspondence above and your local CUDA environment, or fall back to the CPU version.
  • For Windows environments, I personally recommend using the DirectML version.
    // Method to enable GPU
    ModelConfiguration.Instance.SetGPUConfig(GPUBackend.DirectML);