All Downloads are FREE. Search and download functionalities are using the official Maven repository.

org.opencv.dnn.DetectionModel Maven / Gradle / Ivy

The newest version!
//
// This file is auto-generated. Please don't modify it!
//
package org.opencv.dnn;

import java.util.ArrayList;
import java.util.List;
import org.opencv.core.Mat;
import org.opencv.core.MatOfFloat;
import org.opencv.core.MatOfInt;
import org.opencv.core.MatOfRect;
import org.opencv.dnn.DetectionModel;
import org.opencv.dnn.Model;
import org.opencv.dnn.Net;
import org.opencv.utils.Converters;

// C++: class DetectionModel
/**
 * This class represents high-level API for object detection networks.
 *
 * DetectionModel allows to set params for preprocessing input image.
 * DetectionModel creates net from file with trained weights and config,
 * sets preprocessing input, runs forward pass and return result detections.
 * For DetectionModel SSD, Faster R-CNN, YOLO topologies are supported.
 */
public class DetectionModel extends Model {

    protected DetectionModel(long addr) { super(addr); }

    // internal usage only
    public static DetectionModel __fromPtr__(long addr) { return new DetectionModel(addr); }

    //
    // C++:   cv::dnn::DetectionModel::DetectionModel(String model, String config = "")
    //

    /**
     * Create detection model from network represented in one of the supported formats.
     * An order of {@code model} and {@code config} arguments does not matter.
     * @param model Binary file contains trained weights.
     * @param config Text file contains network configuration.
     */
    public DetectionModel(String model, String config) {
        super(DetectionModel_0(model, config));
    }

    /**
     * Create detection model from network represented in one of the supported formats.
     * An order of {@code model} and {@code config} arguments does not matter.
     * @param model Binary file contains trained weights.
     */
    public DetectionModel(String model) {
        super(DetectionModel_1(model));
    }


    //
    // C++:   cv::dnn::DetectionModel::DetectionModel(Net network)
    //

    /**
     * Create model from deep learning network.
     * @param network Net object.
     */
    public DetectionModel(Net network) {
        super(DetectionModel_2(network.nativeObj));
    }


    //
    // C++:  DetectionModel cv::dnn::DetectionModel::setNmsAcrossClasses(bool value)
    //

    /**
     * nmsAcrossClasses defaults to false,
     * such that when non max suppression is used during the detect() function, it will do so per-class.
     * This function allows you to toggle this behaviour.
     * @param value The new value for nmsAcrossClasses
     * @return automatically generated
     */
    public DetectionModel setNmsAcrossClasses(boolean value) {
        return new DetectionModel(setNmsAcrossClasses_0(nativeObj, value));
    }


    //
    // C++:  bool cv::dnn::DetectionModel::getNmsAcrossClasses()
    //

    /**
     * Getter for nmsAcrossClasses. This variable defaults to false,
     * such that when non max suppression is used during the detect() function, it will do so only per-class
     * @return automatically generated
     */
    public boolean getNmsAcrossClasses() {
        return getNmsAcrossClasses_0(nativeObj);
    }


    //
    // C++:  void cv::dnn::DetectionModel::detect(Mat frame, vector_int& classIds, vector_float& confidences, vector_Rect& boxes, float confThreshold = 0.5f, float nmsThreshold = 0.0f)
    //

    /**
     * Given the {@code input} frame, create input blob, run net and return result detections.
     * @param classIds Class indexes in result detection.
     * @param confidences A set of corresponding confidences.
     * @param boxes A set of bounding boxes.
     * @param confThreshold A threshold used to filter boxes by confidences.
     * @param nmsThreshold A threshold used in non maximum suppression.
     * @param frame automatically generated
     */
    public void detect(Mat frame, MatOfInt classIds, MatOfFloat confidences, MatOfRect boxes, float confThreshold, float nmsThreshold) {
        Mat classIds_mat = classIds;
        Mat confidences_mat = confidences;
        Mat boxes_mat = boxes;
        detect_0(nativeObj, frame.nativeObj, classIds_mat.nativeObj, confidences_mat.nativeObj, boxes_mat.nativeObj, confThreshold, nmsThreshold);
    }

    /**
     * Given the {@code input} frame, create input blob, run net and return result detections.
     * @param classIds Class indexes in result detection.
     * @param confidences A set of corresponding confidences.
     * @param boxes A set of bounding boxes.
     * @param confThreshold A threshold used to filter boxes by confidences.
     * @param frame automatically generated
     */
    public void detect(Mat frame, MatOfInt classIds, MatOfFloat confidences, MatOfRect boxes, float confThreshold) {
        Mat classIds_mat = classIds;
        Mat confidences_mat = confidences;
        Mat boxes_mat = boxes;
        detect_1(nativeObj, frame.nativeObj, classIds_mat.nativeObj, confidences_mat.nativeObj, boxes_mat.nativeObj, confThreshold);
    }

    /**
     * Given the {@code input} frame, create input blob, run net and return result detections.
     * @param classIds Class indexes in result detection.
     * @param confidences A set of corresponding confidences.
     * @param boxes A set of bounding boxes.
     * @param frame automatically generated
     */
    public void detect(Mat frame, MatOfInt classIds, MatOfFloat confidences, MatOfRect boxes) {
        Mat classIds_mat = classIds;
        Mat confidences_mat = confidences;
        Mat boxes_mat = boxes;
        detect_2(nativeObj, frame.nativeObj, classIds_mat.nativeObj, confidences_mat.nativeObj, boxes_mat.nativeObj);
    }


    @Override
    protected void finalize() throws Throwable {
        delete(nativeObj);
    }



    // C++:   cv::dnn::DetectionModel::DetectionModel(String model, String config = "")
    private static native long DetectionModel_0(String model, String config);
    private static native long DetectionModel_1(String model);

    // C++:   cv::dnn::DetectionModel::DetectionModel(Net network)
    private static native long DetectionModel_2(long network_nativeObj);

    // C++:  DetectionModel cv::dnn::DetectionModel::setNmsAcrossClasses(bool value)
    private static native long setNmsAcrossClasses_0(long nativeObj, boolean value);

    // C++:  bool cv::dnn::DetectionModel::getNmsAcrossClasses()
    private static native boolean getNmsAcrossClasses_0(long nativeObj);

    // C++:  void cv::dnn::DetectionModel::detect(Mat frame, vector_int& classIds, vector_float& confidences, vector_Rect& boxes, float confThreshold = 0.5f, float nmsThreshold = 0.0f)
    private static native void detect_0(long nativeObj, long frame_nativeObj, long classIds_mat_nativeObj, long confidences_mat_nativeObj, long boxes_mat_nativeObj, float confThreshold, float nmsThreshold);
    private static native void detect_1(long nativeObj, long frame_nativeObj, long classIds_mat_nativeObj, long confidences_mat_nativeObj, long boxes_mat_nativeObj, float confThreshold);
    private static native void detect_2(long nativeObj, long frame_nativeObj, long classIds_mat_nativeObj, long confidences_mat_nativeObj, long boxes_mat_nativeObj);

    // native support for java finalize()
    private static native void delete(long nativeObj);

}




© 2015 - 2024 Weber Informatics LLC | Privacy Policy