Skip to content

Commit 2ea5cea

Browse files
bluehatchbluehatch
committed
Implements face_detection_yunet java demo(opencv#251)
1 parent 0f69e83 commit 2ea5cea

File tree

3 files changed

+199
-0
lines changed

3 files changed

+199
-0
lines changed

‎models/face_detection_yunet/demo.java

+167
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,167 @@
1+
import com.beust.jcommander.JCommander;
2+
import com.beust.jcommander.Parameter;
3+
import com.beust.jcommander.UnixStyleUsageFormatter;
4+
import org.bytedeco.opencv.global.opencv_dnn;
5+
import org.bytedeco.opencv.opencv_core.*;
6+
import org.bytedeco.opencv.opencv_objdetect.FaceDetectorYN;
7+
import org.bytedeco.opencv.opencv_videoio.VideoCapture;
8+
9+
import static org.bytedeco.opencv.global.opencv_highgui.*;
10+
import static org.bytedeco.opencv.global.opencv_imgcodecs.imread;
11+
import static org.bytedeco.opencv.global.opencv_imgcodecs.imwrite;
12+
import static org.bytedeco.opencv.global.opencv_imgproc.FONT_HERSHEY_SIMPLEX;
13+
import static org.bytedeco.opencv.global.opencv_imgproc.putText;
14+
import static org.bytedeco.opencv.global.opencv_videoio.CAP_PROP_FRAME_HEIGHT;
15+
import static org.bytedeco.opencv.global.opencv_videoio.CAP_PROP_FRAME_WIDTH;
16+
17+
public class demo {
18+
19+
// Valid combinations of backends and targets
20+
static int[][] backendTargetPairs = {
21+
{opencv_dnn.DNN_BACKEND_OPENCV, opencv_dnn.DNN_TARGET_CPU},
22+
{opencv_dnn.DNN_BACKEND_CUDA, opencv_dnn.DNN_TARGET_CUDA},
23+
{opencv_dnn.DNN_BACKEND_CUDA, opencv_dnn.DNN_TARGET_CUDA_FP16},
24+
{opencv_dnn.DNN_BACKEND_TIMVX, opencv_dnn.DNN_TARGET_NPU},
25+
{opencv_dnn.DNN_BACKEND_CANN, opencv_dnn.DNN_TARGET_NPU}
26+
};
27+
28+
static class Args {
29+
@Parameter(names = {"--help", "-h"}, order = 0, help = true,
30+
description = "Print help message.")
31+
boolean help;
32+
@Parameter(names = {"--input", "-i"}, order = 1,
33+
description = "Set input to a certain image, omit if using camera.")
34+
String input;
35+
@Parameter(names = {"--model", "-m"}, order = 2,
36+
description = "Set model type.")
37+
String model = "face_detection_yunet_2023mar.onnx";
38+
@Parameter(names = {"--backend_target", "-bt"}, order = 3,
39+
description = "Choose one of the backend-target pair to run this demo:" +
40+
" 0: OpenCV implementation + CPU," +
41+
" 1: CUDA + GPU (CUDA), " +
42+
" 2: CUDA + GPU (CUDA FP16)," +
43+
" 3: TIM-VX + NPU," +
44+
" 4: CANN + NPU")
45+
int backendTarget = 0;
46+
@Parameter(names = {"--conf_threshold"}, order = 5,
47+
description = "Set the minimum needed confidence for the model to identify a face. Filter out faces of conf < conf_threshold")
48+
float confThreshold = 0.9f;
49+
@Parameter(names = {"--nms_threshold"}, order = 5,
50+
description = "Set the threshold to suppress overlapped boxes. Suppress boxes if IoU(box1, box2) >= nms_threshold, the one of higher score is kept.")
51+
float nmsThreshold = 0.3f;
52+
@Parameter(names = {"--top_k"}, order = 5,
53+
description = "Keep top_k bounding boxes before NMS. Set a lower value may help speed up postprocessing.")
54+
int topK = 5000;
55+
@Parameter(names = {"--save", "-s"}, order = 4,
56+
description = "Specify to save file with results (i.e. bounding box, confidence level). Invalid in case of camera input.")
57+
boolean save;
58+
@Parameter(names = {"--vis", "-v"}, order = 5, arity = 1,
59+
description = "Specify to open a new window to show results. Invalid in case of camera input.")
60+
boolean vis = true;
61+
}
62+
63+
64+
static class YuNet {
65+
private final FaceDetectorYN model;
66+
67+
YuNet(String modelPath, Size inputSize, float confThreshold, float nmsThreshold, int topK,
68+
int backendId, int targetId) {
69+
model = FaceDetectorYN.create(modelPath, "", inputSize, confThreshold, nmsThreshold, topK,
70+
backendId, targetId);
71+
}
72+
73+
void setInputSize(Size inputSize) {
74+
model.setInputSize(inputSize);
75+
}
76+
77+
Mat infer(Mat image) {
78+
final Mat res = new Mat();
79+
model.detect(image, res);
80+
return res;
81+
}
82+
}
83+
84+
static Mat visualize(Mat image, Mat faces, Scalar textColor, double fps) {
85+
final Mat output = image.clone();
86+
final int h = output.rows();
87+
final int w = output.cols();
88+
if (fps >= 0) {
89+
putText(output, String.format("FPS: %.2f", fps), new Point(0, 15), FONT_HERSHEY_SIMPLEX, 0.5, textColor);
90+
}
91+
92+
// TODO
93+
return output;
94+
}
95+
96+
public static void main(String[] argv) {
97+
final Args args = new Args();
98+
final JCommander jc = JCommander.newBuilder()
99+
.addObject(args)
100+
.build();
101+
jc.setUsageFormatter(new UnixStyleUsageFormatter(jc));
102+
jc.parse(argv);
103+
if (args.help) {
104+
jc.usage();
105+
return;
106+
}
107+
final int backendId = backendTargetPairs[args.backendTarget][0];
108+
final int targetId = backendTargetPairs[args.backendTarget][1];
109+
final YuNet model = new YuNet(args.model, new Size(320, 320), args.confThreshold, args.nmsThreshold,
110+
args.topK, backendId, targetId);
111+
112+
if (args.input != null) {
113+
final Mat image = imread(args.input);
114+
115+
// Inference
116+
model.setInputSize(image.size());
117+
final Mat faces = model.infer(image);
118+
119+
// Print faces
120+
System.out.printf("%d faces detected:\n", faces.rows());
121+
for (int i = 0; i < faces.rows(); i++) {
122+
// TODO
123+
}
124+
125+
// Draw reults on the input image
126+
if (args.save || args.vis) {
127+
final Mat resImage = visualize(image, faces, new Scalar(0, 0, 255, 0), -1);
128+
if (args.save) {
129+
System.out.println("Results are saved to result.jpg");
130+
imwrite("result.jpg", resImage);
131+
}
132+
if (args.vis) {
133+
namedWindow(args.input, WINDOW_AUTOSIZE);
134+
imshow(args.input, resImage);
135+
waitKey(0);
136+
}
137+
}
138+
139+
} else { // // Call default camera
140+
final int deviceId = 0;
141+
final VideoCapture cap = new VideoCapture(deviceId);
142+
final int w = (int) cap.get(CAP_PROP_FRAME_WIDTH);
143+
final int h = (int) cap.get(CAP_PROP_FRAME_HEIGHT);
144+
model.setInputSize(new Size(w, h));
145+
146+
final TickMeter tm = new TickMeter();
147+
final Mat frame = new Mat();
148+
while (waitKey(1) < 0) {
149+
boolean hasFrame = cap.read(frame);
150+
if (!hasFrame) {
151+
System.out.println("No frames grabbed! Exiting ...");
152+
break;
153+
}
154+
// Inference
155+
tm.start();
156+
final Mat faces = model.infer(frame);
157+
tm.stop();
158+
final Mat resImage = visualize(frame, faces, new Scalar(0, 0, 255, 0), tm.getFPS());
159+
imshow("YuNet Demo", resImage);
160+
161+
tm.reset();
162+
}
163+
}
164+
}
165+
166+
167+
}

‎models/face_detection_yunet/pom.xml

+31
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,31 @@
1+
<?xml version="1.0" encoding="UTF-8"?>
2+
3+
<project xmlns="http://maven.apache.org/POM/4.0.0"
4+
xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
5+
xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 https://maven.apache.org/xsd/maven-4.0.0.xsd">
6+
<modelVersion>4.0.0</modelVersion>
7+
8+
<parent>
9+
<groupId>opencv_zoo</groupId>
10+
<artifactId>demo</artifactId>
11+
<version>1.0.0-SNAPSHOT</version>
12+
</parent>
13+
14+
<artifactId>face_detection_yunet</artifactId>
15+
16+
<build>
17+
<sourceDirectory>${project.basedir}</sourceDirectory>
18+
<plugins>
19+
<plugin>
20+
<groupId>org.codehaus.mojo</groupId>
21+
<artifactId>exec-maven-plugin</artifactId>
22+
<version>3.3.0</version>
23+
<configuration>
24+
<executable>java</executable>
25+
<mainClass>demo</mainClass>
26+
</configuration>
27+
</plugin>
28+
</plugins>
29+
</build>
30+
31+
</project>

‎models/pom.xml

+1
Original file line numberDiff line numberDiff line change
@@ -27,6 +27,7 @@
2727
</build>
2828

2929
<modules>
30+
<module>face_detection_yunet</module>
3031
<module>object_tracking_vittrack</module>
3132
<module>text_detection_ppocr</module>
3233
</modules>

0 commit comments

Comments
 (0)