Merge pull request #1828 from shimat/FaceDetectorYN_test

Add tests for FaceDetectorYN

Author: shimat

test/OpenCvSharp.Tests/objdetect/FaceDetectorYNTest.cs

@@ -0,0 +1,170 @@
+using OpenCvSharp.Dnn;
+using Xunit;
+
+namespace OpenCvSharp.Tests.ObjDetect;
+
+// ReSharper disable InconsistentNaming
+public class FaceDetectorYNTest : TestBase
+{
+    // YuNet face detection model from OpenCV Zoo
+    // https://github.com/opencv/opencv_zoo/tree/main/models/face_detection_yunet
+    private const string ModelUrl =
+        "https://github.com/opencv/opencv_zoo/raw/main/models/face_detection_yunet/face_detection_yunet_2023mar.onnx";
+
+    private const string ModelPath = "_data/model/face_detection_yunet_2023mar.onnx";
+
+    static FaceDetectorYNTest()
+    {
+        if (!File.Exists(ModelPath))
+        {
+            var contents = FileDownloader.DownloadData(new Uri(ModelUrl));
+            File.WriteAllBytes(ModelPath, contents);
+        }
+    }
+
+    [Fact]
+    public void Create()
+    {
+        using var detector = new FaceDetectorYN(
+            ModelPath,
+            config: "",
+            inputSize: new Size(320, 320));
+
+        Assert.NotNull(detector);
+    }
+
+    [Fact]
+    public void CreateWithParameters()
+    {
+        using var detector = new FaceDetectorYN(
+            ModelPath,
+            config: "",
+            inputSize: new Size(640, 480),
+            scoreThreshold: 0.8f,
+            nmsThreshold: 0.4f,
+            topK: 3000,
+            backendId: Backend.DEFAULT,
+            targetId: Target.CPU);
+
+        Assert.NotNull(detector);
+    }
+
+    [Fact]
+    public void Dispose_Twice()
+    {
+        var detector = new FaceDetectorYN(
+            ModelPath,
+            config: "",
+            inputSize: new Size(320, 320));
+
+        detector.Dispose();
+        detector.Dispose(); // Should not throw
+    }
+
+    [Fact]
+    public void Detect_Lenna()
+    {
+        using var image = LoadImage("lenna.png");
+        Assert.False(image.Empty());
+
+        using var detector = new FaceDetectorYN(
+            ModelPath,
+            config: "",
+            inputSize: new Size(image.Cols, image.Rows),
+            scoreThreshold: 0.9f,
+            nmsThreshold: 0.3f,
+            topK: 5000);
+
+        using var faces = new Mat();
+        int result = detector.Detect(image, faces);
+
+        // Lenna image should contain at least one face
+        Assert.Equal(1, result);
+        Assert.False(faces.Empty());
+
+        // Each row in faces is a detected face:
+        // [x, y, w, h, x_re, y_re, x_le, y_le, x_nt, y_nt, x_rcm, y_rcm, x_lcm, y_lcm, score]
+        // 14 landmarks + 1 score = 15 columns
+        Assert.True(faces.Rows > 0, "Expected at least one detected face");
+        Assert.Equal(15, faces.Cols);
+
+        // Verify the confidence score (last column) is reasonable
+        var score = faces.At<float>(0, 14);
+        Assert.True(score > 0.5f, $"Face confidence score {score} is unexpectedly low");
+
+        ShowImagesWhenDebugMode(image);
+    }
+
+    [Fact]
+    public void Detect_NoFace()
+    {
+        // Create a blank image with no face
+        using var image = new Mat(320, 320, MatType.CV_8UC3, Scalar.All(128));
+
+        using var detector = new FaceDetectorYN(
+            ModelPath,
+            config: "",
+            inputSize: new Size(320, 320),
+            scoreThreshold: 0.9f);
+
+        using var faces = new Mat();
+        detector.Detect(image, faces);
+
+        // Blank image should have no detected faces
+        Assert.True(faces.Empty() || faces.Rows == 0,
+            $"Expected no faces but got {faces.Rows} detection(s)");
+    }
+
+    [Fact]
+    public void Detect_MultipleCalls()
+    {
+        using var image = LoadImage("lenna.png");
+        using var detector = new FaceDetectorYN(
+            ModelPath,
+            config: "",
+            inputSize: new Size(image.Cols, image.Rows));
+
+        // Verify that calling Detect multiple times produces consistent results
+        using var faces1 = new Mat();
+        using var faces2 = new Mat();
+
+        int result1 = detector.Detect(image, faces1);
+        int result2 = detector.Detect(image, faces2);
+
+        Assert.Equal(result1, result2);
+        Assert.Equal(faces1.Rows, faces2.Rows);
+        Assert.Equal(faces1.Cols, faces2.Cols);
+    }
+
+    [Fact]
+    public void Detect_DifferentThresholds()
+    {
+        using var image = LoadImage("lenna.png");
+
+        // With high threshold
+        using var detectorHigh = new FaceDetectorYN(
+            ModelPath,
+            config: "",
+            inputSize: new Size(image.Cols, image.Rows),
+            scoreThreshold: 0.99f);
+
+        using var facesHigh = new Mat();
+        detectorHigh.Detect(image, facesHigh);
+
+        // With low threshold
+        using var detectorLow = new FaceDetectorYN(
+            ModelPath,
+            config: "",
+            inputSize: new Size(image.Cols, image.Rows),
+            scoreThreshold: 0.1f);
+
+        using var facesLow = new Mat();
+        detectorLow.Detect(image, facesLow);
+
+        // Lower threshold should yield >= detections than higher threshold
+        int highCount = facesHigh.Empty() ? 0 : facesHigh.Rows;
+        int lowCount = facesLow.Empty() ? 0 : facesLow.Rows;
+        Assert.True(lowCount >= highCount,
+            $"Low threshold detections ({lowCount}) should be >= high threshold detections ({highCount})");
+    }
+}