0
Ich bin neu zu diesem Thema. Ich versuche viele SVMs für den gleichen Datensatz mit verschiedenen Parametern zu trainieren (verschiedene Arten von Bildern von Bildern in der Zukunft, hoffe ich), um dann jede SVM vorherzusagen und die meist gefundene Klasse zu akzeptieren.Ich kann SVM nicht wie erwartet arbeiten
Ich habe versucht, den Code vieler Leute über SVM-Bildtraining zu lesen, konnte aber nicht herausfinden, was ich falsch mache, in meinem Code. Was auch immer ich versuche, svm.Predict immer 0.
Jede Hilfe oder Hinweis wird sehr geschätzt.
internal class SVMClassifier
{
Dictionary<int, string> classIndex_name;
List<SVM> svms;
internal void Train(string trainFolder)
{
this.classIndex_name = new Dictionary<int, string>();
Dictionary<int, List<Mat>> class_mats = getMats(trainFolder, this.classIndex_name);
this.svms = new List<SVM>();
Mat samples; Mat responses;
getTrainingData(class_mats, out samples, out responses);
svms.Add(trainSVM(samples, responses));
svms.Add(trainSVM(samples, responses, SVM.SvmType.CSvc, SVM.SvmKernelType.Linear, 0d, 0d, 10d, TermCritType.Iter | TermCritType.Eps, 1000, 0.000001d, 0d, 0d));
svms.Add(trainSVM(samples, responses, SVM.SvmType.CSvc, SVM.SvmKernelType.Rbf, 100d, 100d, 1d, TermCritType.Iter | TermCritType.Eps, 1000, 0.000001d, 0.1d, 0.5d));
samples.Dispose(); responses.Dispose();
foreach (Mat mat in class_mats.Values.SelectMany((a) => a))
mat.Dispose();
}
private static Dictionary<int, List<Mat>> getMats(string trainFolder, Dictionary<int, string> classIndex_name)
{
Dictionary<int, List<Mat>> class_mats = new Dictionary<int, List<Mat>>();
DirectoryInfo diTrain = new DirectoryInfo(trainFolder);
int i = 0;
foreach (var di in diTrain.GetDirectories())//classes are according to the directories
{
var dirName = di.Name;
classIndex_name[i] = dirName;
var fileNames = di.GetFiles().Select((a) => a.FullName).ToList();
fileNames.Sort(new Dece.Misc.NumericSuffixFileFullNameComparer());
class_mats[i] = fileNames.Select((a) => getMat(a, true)).ToList();
i++;
}
return class_mats;
}
private static SVM trainSVM(Mat samples, Mat responses,
SVM.SvmType? svm_Type = null, SVM.SvmKernelType? svm_KernelType = null, double? gamma = null, double? degree = null, double? c = null,
TermCritType? criteriaType = null, int? criteriaMaxCount = null, double? criteriaEps = null, double? p = null, double? nu=null)
{
SVM svm = new SVM();
if (svm_Type != null) svm.Type = (SVM.SvmType)svm_Type;
if (svm_KernelType != null) svm.SetKernel((SVM.SvmKernelType)svm_KernelType);
if (gamma != null) svm.Gamma = (double)gamma;
if (degree != null) svm.Degree = (double)degree;
if (c != null) svm.C = (double)c;
if ((criteriaType != null) || (criteriaMaxCount != null) || (criteriaEps != null))
{
var t = new MCvTermCriteria((int)criteriaMaxCount, (double)criteriaEps);
if (criteriaType != null) t.Type = (TermCritType)criteriaType;
svm.TermCriteria = t;
}
if (p != null) svm.P = (double)p;
if (nu != null) svm.Nu = (double)nu;
if (!svm.Train(samples, DataLayoutType.RowSample, responses))
throw new Exception();
return svm;
}
private static void getTrainingData(Dictionary<int, List<Mat>> class_mats, out Mat samples, out Mat responses)
{
samples = null;
List<int> lstResp = new List<int>();
foreach (int cls in class_mats.Keys)
{
int count = 0;
foreach (Mat mat in class_mats[cls])
using (var desc = mat.Reshape(0, 1))
{
if (samples == null)
samples = new Mat(desc.Cols, 0, desc.Depth, 1);
samples.PushBack(desc);
count += desc.Rows;
}
for (int i = 0; i < count; i++)
lstResp.Add(cls);
}
//responses = new Mat(new Size(lstResp.Count, 1), DepthType.Cv32S, 1);
//for (int i = 0; i < lstResp.Count; i++)
// responses.SetValue(0, i, lstResp[i]);
responses = new Mat(new Size(1, lstResp.Count), DepthType.Cv32S, 1);
for (int i = 0; i < lstResp.Count; i++)
responses.SetValue(i, 0, lstResp[i]);
if (samples.Depth != DepthType.Cv32F)
samples.ConvertTo(samples, DepthType.Cv32F);
CvInvoke.Normalize(samples, samples, -1, 1, NormType.MinMax);
}
internal void Detect(IEnumerable<string> fileNames, Action<ShapeInfo> detected)
{
foreach (var fn in fileNames)
using (Mat mat = getMat(fn, false))
{
{
using (var samples = mat.Reshape(0, 1))
{
if (samples.Depth != DepthType.Cv32F)
samples.ConvertTo(samples, DepthType.Cv32F);
CvInvoke.Normalize(samples, samples, -1, 1, NormType.MinMax);
foreach (var svm in this.svms)
{
Mat res = new Mat();
float p0 = svm.Predict(samples, res, 0);
float p1 = svm.Predict(samples, res, 1);
float p2 = svm.Predict(samples, res, 2);
float p3 = svm.Predict(samples, res, 3);
float p4 = svm.Predict(samples, res, 4);
float p = svm.Predict(samples, res);
foreach (var val in toIEnumerable(p0, p1, p2, p3, p4, p))
if (val != 0f)
{
System.Windows.Forms.MessageBox.Show("never enters here :(");
}
}
}
}
}
}
private static Mat getMat(string fn, bool train)
{
var mat = new Mat(fn, ImreadModes.Grayscale);
mat.Resize(new Size(128, 128));
return mat;
}
private static IEnumerable<T> toIEnumerable<T>(params T[] items)
{
if (items != null)
foreach (var item in items)
yield return item;
}
}
Mat.SetValue Erweiterung von here genommen.
Ich hoffe zu fragen, wie das Format dieser Website geeignet ist. Wenn diese Frage nicht geschlossen werden kann, ist das kein Problem. Ich versuche zu verstehen, wie wir ein SVM mit Bildern trainieren sollten.