Revision history [back]

Im trying to do 3 class classification with 5 samples for each class on android But im getting this error OpenCV Error: Assertion failed (nsamples == 1) in virtual float cv::ml::SVMImpl::predict ?

Here is my code

public ImageDetection(Context mContext) throws IOException {
        context = mContext;
        TermCriteria tc = new TermCriteria(CV_TERMCRIT_ITER, 10, 0.001);
        dictionarySize = 1500;
        int retries = 1;
        bowTrainer = new BOWKMeansTrainer(dictionarySize, tc, retries, KMEANS_PP_CENTERS);

        descriptorMatcher = DescriptorMatcher.create(DescriptorMatcher.FLANNBASED);
        bowImgDescriptorExtractor = new BOWImgDescriptorExtractor(descriptorExtractor, descriptorMatcher);

        trainDescriptor = new MatVector();
        evalDescriptor = new Mat();
        trainLabs = new int[15];
        evalLabs = new int[15];

        initializeSVM();
        collectClassCentroids();
        startClustering();
        evaluate();

    }

    private void initializeSVM() {
        svm.setKernel(SVM.RBF);
        svm.setType(SVM.C_SVC);
        svm.setGamma(0.50625000000000009);
        svm.setC(312.50000000000000);
        svm.setTermCriteria(new TermCriteria(CV_TERMCRIT_ITER, 100, 0.000001));
    }

    private void collectClassCentroids() throws IOException {
        int[][] train = new int[][]{{R.drawable.train1_1, R.drawable.train1_2, R.drawable.train1_3, R.drawable.train1_4, R.drawable.train1_5},
                {R.drawable.train2_1, R.drawable.train2_2, R.drawable.train2_3, R.drawable.train2_4, R.drawable.train2_5},
                {R.drawable.train3_1, R.drawable.train3_2, R.drawable.train3_3, R.drawable.train3_4, R.drawable.train3_5}};

        int counter1 = 0;
        for (int i = 0; i < 3; i++) {
            for (int j = 0; j < 5; j++) {
                Mat src = loadResource(train[i][j]);
                if (!src.empty()) {
                    cvtColor(src,src, CV_RGBA2GRAY);
                    KeyPointVector mKeypoint = new KeyPointVector();
                    featureDetector.detect(src, mKeypoint);
                    Mat mFeatures = new Mat();
                    descriptorExtractor.compute(src, mKeypoint, mFeatures);
                    trainLabs[counter1] = i;
                    counter1++;
                    bowTrainer.add(mFeatures);
                }
            }
        }

    }

    public void startClustering() throws IOException {

        trainDescriptor = bowTrainer.getDescriptors();
        int countRow = 0;
        for (int i = 0; i < trainDescriptor.size(); i++) {
            countRow += trainDescriptor.get(i).rows();
        }

        Mat dictionary = bowTrainer.cluster();
        bowImgDescriptorExtractor.setVocabulary(dictionary);

        trainLabel = new Mat(trainLabs);

        Mat trainingData = new Mat(0, dictionarySize, CV_32FC1);

        KeyPointVector keyPointVector1 = new KeyPointVector();
        Mat bowDescriptor1 = new Mat();

        int[][] train = new int[][]{{R.drawable.train1_1, R.drawable.train1_2, R.drawable.train1_3, R.drawable.train1_4, R.drawable.train1_5},
                {R.drawable.train2_1, R.drawable.train2_2, R.drawable.train2_3, R.drawable.train2_4, R.drawable.train2_5},
                {R.drawable.train3_1, R.drawable.train3_2, R.drawable.train3_3, R.drawable.train3_4, R.drawable.train3_5}};

        for (int i = 0; i < 3; i++) {
            for (int j = 0; j < 5; j++) {
                Mat src = loadResource(train[i][j]);
                cvtColor(src, src, CV_RGBA2GRAY);
                featureDetector.detect(src, keyPointVector1);
                bowImgDescriptorExtractor.compute(src, keyPointVector1, bowDescriptor1);
                trainingData.push_back(bowDescriptor1);
            }
        }

        System.out.println("Training Data "+trainingData.rows()+","+trainingData.cols());
        System.out.println("Training Label "+trainLabel.rows()+","+trainLabel.cols());
        svm.train(trainingData, ROW_SAMPLE, trainLabel);
    }

    public void evaluate() throws IOException {

        int [] evalLabel = trainLabs;
        Mat groundTruth = new Mat(evalLabel);
        Mat evalData = new Mat(0,dictionarySize,CV_32FC1);
        KeyPointVector keyPointVectorEval = new KeyPointVector();
        Mat bowDescriptor = new Mat();

        Mat results = new Mat(0,1,CV_32FC1);

        int[][] eval = new int[][]{{R.drawable.train1_1, R.drawable.train1_2, R.drawable.train1_3, R.drawable.train1_4, R.drawable.train1_5},
                {R.drawable.train2_1, R.drawable.train2_2, R.drawable.train2_3, R.drawable.train2_4, R.drawable.train2_5},
                {R.drawable.train3_1, R.drawable.train3_2, R.drawable.train3_3, R.drawable.train3_4, R.drawable.train3_5}};

        for (int i = 0; i < 4; 3; i++) {
            for (int j = 0; j <5 ; j++) {
                Mat rImg = loadResource(eval[i][j]);
                cvtColor(rImg, rImg, CV_RGBA2GRAY);
                featureDetector.detect(rImg, keyPointVectorEval);
                bowImgDescriptorExtractor.compute(rImg, keyPointVectorEval, bowDescriptor);

                //evalData.push_back(bowDescriptor);
                System.out.print("Ground Truth : "+evalLabel[i]);
                Mat pass = bowDescriptor.reshape(1,1);
                evalData.push_back(pass);
                float respons = svm.predict(evalData);
                int getInt = ((int) respons);
                System.out.println("Response : "+respons);
            }
        }


       // Mat subs = subtractPut(groundTruth, results);
        //double errorRate = (double)countNonZero(subs) / evalData.rows();
        //System.out.println("Error rate is : "+errorRate);

    }

    private Mat loadResource(int resourceId) throws IOException {
        InputStream is = context.getResources().openRawResource(resourceId);
        int bytesRead;
        byte[] data = new byte[16 * 1024];
        ByteArrayOutputStream buffer = new ByteArrayOutputStream();
        while ((bytesRead = is.read(data, 0, data.length)) != -1) {
            buffer.write(data, 0, bytesRead);
        }
        byte[] bytes = buffer.toByteArray();
        Mat mat = imdecode(new Mat(bytes), CV_32FC1);
        return mat;
    }