[捉虫记录]关于Cascade Training Error的bug

Jeremy 

opencv_traincascade.exe -data data -vec diode\pos.vec -bg nodiode\neg.dat -numPos 600 -numNeg 600 -numStages 15 -w 22 -h 52 -minHitRate 0.995 -maxFalseAlarmRate 0.5 
pause

Note:如果上面的参数名称写错了,程序会自动忽略,并取默认数值.

在应用opencv_traincascade.exe对图片进行训练的过程中碰到很多问题,现一一记录如下:

[bug1]: Train dataset for temp stage can not filled. Branch training terminated.

首先我们来看这个错误时从哪里出来的,从源代码[cascadeclassifier.cpp]中我们可以看到:

    for( int i = startNumStages; i < numStages; i++ )
    {
        cout << endl << "===== TRAINING " << i << "-stage =====" << endl;
        cout << "<BEGIN" << endl;

        if ( !updateTrainingSet( requiredLeafFARate, tempLeafFARate ) )
        {
            cout << "Train dataset for temp stage can not be filled. "
                "Branch training terminated." << endl;
            break;
        }
        if( tempLeafFARate <= requiredLeafFARate )
        {
            cout << "Required leaf false alarm rate achieved. "
                 "Branch training terminated." << endl;
            break;
        }

        CvCascadeBoost* tempStage = new CvCascadeBoost;
        bool isStageTrained = tempStage->train( (CvFeatureEvaluator*)featureEvaluator,
                                                curNumSamples, _precalcValBufSize, _precalcIdxBufSize,
                                                *((CvCascadeBoostParams*)stageParams) );
        cout << "END>" << endl;

可以知道,是在函数updateTrainingSet()中出了问题,我们接下来进去看看:

bool CvCascadeClassifier::updateTrainingSet( double minimumAcceptanceRatio, double& acceptanceRatio)
{
    int64 posConsumed = 0, negConsumed = 0;
    imgReader.restart();
    int posCount = fillPassedSamples( 0, numPos, true, 0, posConsumed );
    if( !posCount )
        return false;
    cout << "POS count : consumed   " << posCount << " : " << (int)posConsumed << endl;

    int proNumNeg = cvRound( ( ((double)numNeg) * ((double)posCount) ) / numPos ); // apply only a fraction of negative samples. double is required since overflow is possible
    int negCount = fillPassedSamples( posCount, proNumNeg, false, minimumAcceptanceRatio, negConsumed );
    if ( !negCount )
        return false;

    curNumSamples = posCount + negCount;
    acceptanceRatio = negConsumed == 0 ? 0 : ( (double)negCount/(double)(int64)negConsumed );
    cout << "NEG count : acceptanceRatio    " << negCount << " : " << acceptanceRatio << endl;
    return true;
}

我们的错误输出是在"POS count……"之后，“NEG count”之前，这样……问题就是negCount=false，也即fillPassedSamples（）函数出错了

后来我把numPos变大为1500，numNeg变大为1200。因为每次stage的迭代过程中会只更换“1%”左右的样本。