javacpp :内存使用率太高: physicalBytes > maxPhysicalBytes javacpp指针释放程序()

Question

我有这样的问题：

org.bytedeco.javacpp.Pointer.deallocator(Pointer.java:562)的物理内存使用率太高: physicalBytes = 1G > maxPhysicalBytes = 1G

long Pointer.physicalBytes正在递增，即使我们释放每个指针对象并调用GC --我一直在监视JVM堆大小，并且它处于控制之下，从来没有超过20%的使用率，这意味着解除分配执行得很好，但出于某种原因，信息(真正的信息)没有传递给Poniter.physicalBytes (它永远不会减少)，当错误到达Pointer.maxPhysicalBytes值时，它会错误地抛出错误。

看起来这是几周前修复的(https://github.com/bytedeco/javacpp-presets/issues/423)，但即使在获得最新版本的JavaCPP (1.3.3)之后，我仍然有这个问题。

这是我的密码：

import static org.bytedeco.javacpp.opencv_core.cvClearMemStorage;
import static org.bytedeco.javacpp.opencv_core.cvGetSeqElem;
import static org.bytedeco.javacpp.opencv_core.cvPoint;
import static org.bytedeco.javacpp.opencv_core.cvSize;
import static org.bytedeco.javacpp.opencv_imgproc.CV_AA;
import static org.bytedeco.javacpp.opencv_imgproc.cvRectangle;
import static org.bytedeco.javacpp.opencv_objdetect.cvHaarDetectObjects;

import org.bytedeco.javacpp.BytePointer;
import org.bytedeco.javacpp.Pointer;
import org.bytedeco.javacpp.opencv_core;
import org.bytedeco.javacpp.opencv_core.CvMemStorage;
import org.bytedeco.javacpp.opencv_core.CvPoint;
import org.bytedeco.javacpp.opencv_core.CvRect;
import org.bytedeco.javacpp.opencv_core.CvScalar;
import org.bytedeco.javacpp.opencv_core.CvSeq;
import org.bytedeco.javacpp.opencv_core.CvSize;
import org.bytedeco.javacpp.opencv_core.IplImage;
import org.bytedeco.javacpp.opencv_objdetect.CascadeClassifier;
import org.bytedeco.javacpp.opencv_objdetect.CvHaarClassifierCascade;
(...)    

public class ObjectDetection {

    private static CvMemStorage storage = CvMemStorage.create();

    (...)

    public static synchronized Detection detect(IplImage src, Configuration cfg) {

        CvMemStorage storage = CvMemStorage.create();
        CvSeq sign = cvHaarDetectObjects(src, cfg.cascade, storage, cfg.scale, cfg.neighbors, cfg.method.getVal(), cfg.minSize, cfg.maxSize);

        int total_objs = sign.total();

        for (int i = 0; i < total_objs; i++) {
            BytePointer seqElem = cvGetSeqElem(sign, i);
            CvRect r = new CvRect(seqElem);
            CvPoint p1 = cvPoint(r.x(), r.y());
            CvPoint p2 = cvPoint(r.width() + r.x(), r.height() + r.y());
            cvRectangle(src, p1, p2, CvScalar.RED, 2, CV_AA, 0);
            p1.deallocate();
            p2.deallocate();
            r.close();
            r.deallocate();
            seqElem.deallocate();
        }

        BufferedImage img = Images.toBufferedImage(src);

        sign.deallocate();
        src.deallocate();
        storage.deallocate();

        Pointer.deallocateReferences();

        return new Detection(img, total_objs);
    }

    public static class Detection{
        private BufferedImage img;
        private int count;
        private Detection(BufferedImage i, int c){
            img = i; count = c;
        }
        public BufferedImage getImage(){
            return img;
        }
        public int getObjectsCount(){
            return count;
        }
    }


    public static class Configuration{
        private String configName;
        private CascadeClassifier xmlFile;
        private CvHaarClassifierCascade cascade;
        private CvSize minSize; 
        private CvSize maxSize;
        private double scale;
        private int neighbors;
        private Method method;

        private Configuration(String configuration) throws JSONException, IOException{
            configName = configuration;
            JSONObject cfg = new JSONObject(new JSONTokener(new FileReader(new File(configuration+".cfg"))));
            scale = cfg.getDouble("scale");
            neighbors = cfg.getInt("neighbors");
            method = Method.valueOf(cfg.getString("method"));
            int min = cfg.getInt("min_size");
            int max = cfg.getInt("max_size");
            minSize = cvSize(min,min);
            maxSize = cvSize(max,max);
            xmlFile = new CascadeClassifier(configuration+".xml");
            cascade = new CvHaarClassifierCascade(xmlFile.getOldCascade());
        }

        public void dealocate(){
            xmlFile.deallocate();
            cascade.deallocate();
            minSize.deallocate();
            maxSize.deallocate();
            configs.remove(configName);
        }

    }
}

这就是堆栈轨迹：

java.lang.OutOfMemoryError: Physical memory usage is too high: physicalBytes = 1G > maxPhysicalBytes = 1G
org.bytedeco.javacpp.Pointer.deallocator(Pointer.java:576)
org.bytedeco.javacpp.Pointer.init(Pointer.java:121)
org.bytedeco.javacpp.opencv_core.cvPoint(Native Method)
br.com.irisbot.visualrecognition.haarcascade.ObjectDetection.detect(ObjectDetection.java:83)
br.com.irisbot.visualrecognition.haarcascade.ObjectDetectionServlet.doPost(ObjectDetectionServlet.java:71)
javax.servlet.http.HttpServlet.service(HttpServlet.java:660)
javax.servlet.http.HttpServlet.service(HttpServlet.java:741)
org.apache.catalina.core.ApplicationFilterChain.internalDoFilter(ApplicationFilterChain.java:231)
org.apache.catalina.core.ApplicationFilterChain.doFilter(ApplicationFilterChain.java:166)
org.apache.tomcat.websocket.server.WsFilter.doFilter(WsFilter.java:53)
org.apache.catalina.core.ApplicationFilterChain.internalDoFilter(ApplicationFilterChain.java:193)
org.apache.catalina.core.ApplicationFilterChain.doFilter(ApplicationFilterChain.java:166)
org.apache.catalina.core.StandardWrapperValve.invoke(StandardWrapperValve.java:199)
org.apache.catalina.core.StandardContextValve.invoke(StandardContextValve.java:96)
org.apache.catalina.authenticator.AuthenticatorBase.invoke(AuthenticatorBase.java:475)
org.apache.catalina.core.StandardHostValve.invoke(StandardHostValve.java:140)
org.apache.catalina.valves.ErrorReportValve.invoke(ErrorReportValve.java:81)
org.apache.catalina.valves.AbstractAccessLogValve.invoke(AbstractAccessLogValve.java:651)
org.apache.catalina.core.StandardEngineValve.invoke(StandardEngineValve.java:87)
org.apache.catalina.connector.CoyoteAdapter.service(CoyoteAdapter.java:342)
org.apache.coyote.http11.Http11Processor.service(Http11Processor.java:500)
org.apache.coyote.AbstractProcessorLight.process(AbstractProcessorLight.java:66)
org.apache.coyote.AbstractProtocol$ConnectionHandler.process(AbstractProtocol.java:754)
org.apache.tomcat.util.net.NioEndpoint$SocketProcessor.doRun(NioEndpoint.java:1376)
org.apache.tomcat.util.net.SocketProcessorBase.run(SocketProcessorBase.java:49)
java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
org.apache.tomcat.util.threads.TaskThread$WrappingRunnable.run(TaskThread.java:61)
java.lang.Thread.run(Thread.java:748)

Answer 1

事实上，经过很长一段时间来解决内存问题，以及对生产服务的严重需求(在某些请求之后不挂起)，我解决了这个问题：

1. 创建一个独立应用程序(runnable )来执行OpenCV过程，并将结果返回到控制台
2. 创建一个web服务，将应用程序作为Java进程调用到OS命令行中(这将强制它在单独的JVM中运行)
3. 捕获进程控制台的结果
4. 通过web服务返回结果

我同意，你可能会觉得它不那么优雅，但到了一天结束的时候，它会起作用的！一点也不差;-)

Answer 2

很可能有些对象没有得到正确的分配。尝试使用PointerScope来更容易地捕获所有的人：http://bytedeco.org/news/2018/07/17/bytedeco-as-distribution/

下面是一个演示如何与CascadeClassifier一起使用它的最小示例：

import org.bytedeco.javacpp.*;
import org.bytedeco.javacv.*;
import static org.bytedeco.javacpp.opencv_core.*;
import static org.bytedeco.javacpp.opencv_imgproc.*;
import static org.bytedeco.javacpp.opencv_objdetect.*;

public class PointerScopeDemo {
    public static void main(String[] args) throws Exception {
        CascadeClassifier classifier = new CascadeClassifier("haarcascade_frontalface_alt2.xml");
        FrameGrabber grabber = FrameGrabber.createDefault(0);
        grabber.start();

        Mat image;
        OpenCVFrameConverter.ToMat converter = new OpenCVFrameConverter.ToMat();
        while ((image = converter.convert(grabber.grab())) != null) {
            try (PointerScope scope = new PointerScope()) {
                RectVector faces = new RectVector();
                classifier.detectMultiScale(image, faces);
                System.out.println(faces.size());
            }
        }
    }
}