/*
* for each prediction, this method create a line following this schema:
* {image_name} {label} {score}
*/
private List <string> ExportClassificationResult(TomatoClassifier classifier, List <Tomato> testTomatoes)
{
var classifiedResult = new List <string>();
log.LogNewLine("compute label on test images..");
foreach (Tomato tomato in testTomatoes)
{
var score = 0.0;
var res = classifier.Classify(tomato.ColoredImage, tomato.GrayImage1, tomato.GrayImage2, out score);
classifiedResult.Add(tomato.Name + " " + (int)res + " " + score);
}
return(classifiedResult);
}
public void CheckPerformance()
{
this.images.Clear();
View.ClearImages();
log.LogNewLine("eval starts..");
Train(classifier, tomatoSets[SetType.TRANING]);
foreach (Tomato tomato in tomatoSets[SetType.VALIDATION])
{
log.Log(".");
double score = 0;
var result = classifier.Classify(tomato.ColoredImage, tomato.GrayImage1, tomato.GrayImage2, out score);
if (result != (int)tomato.Type)
{
this.images.Add(new VisualEvent("WRONG! expeted : " + tomato.Type, tomato.ColoredImage.ToBitmap()));
//this makes process extraction visualization possible:
imageExtractor.ExtractDescriptor(tomato.ColoredImage, tomato.GrayImage1, tomato.GrayImage2, (int)tomato.Type);
}
}
log.LogNewLine("");
View.AddImages(this.images);
}
/*
* this method eval the classifier performance in the validation set.
* the metrics used are:
* - precision;
* - recall;
* - accuracy;
* - confusion matrix.
*/
private void EvalPerformance(TomatoClassifier classifier, List <Tomato> tomatoes)
{
var falsePositive = 0.0;
var falseNegative = 0.0;
var truePositive = 0.0;
var trueNegative = 0.0;
log.LogNewLine("eval...");
stopWatch.Restart();
foreach (Tomato tomato in tomatoes)
{
log.Log(".");
double score = 0;
var predicted = classifier.Classify(tomato.ColoredImage, tomato.GrayImage1, tomato.GrayImage2, out score);
if (predicted == (int)tomato.Type)
{
var updated = predicted == (int)(TomatoType.BAD) ? trueNegative++ : truePositive++;
}
else
{
var updated = predicted == (int)(TomatoType.BAD) ? falseNegative++ : falsePositive++;
}
}
stopWatch.Stop();
log.LogNewLine("");
log.LogNewLine("predict " + tomatoes.Count / stopWatch.Elapsed.Seconds + " tomatoes per second");
log.LogNewLine("ends..");
log.LogNewLine("precision : " + truePositive / (truePositive + falsePositive));
log.LogNewLine("recall : " + truePositive / (truePositive + falseNegative));
log.LogNewLine("accuracy : " + (truePositive + trueNegative) / tomatoes.Count);
log.LogNewLine("-- CONFUSION MATRIX --");
log.LogNewLine("BAD\t" + trueNegative + "\t" + falseNegative);
log.LogNewLine("OK \t" + falsePositive + "\t" + truePositive);
log.LogNewLine(" \tBAD\tOK");
log.LogNewLine("...");
}