static void getTestBenchamrkRes(CaffeModel model, string iscorrect)
{
string fatherDir = @"C:\Data\Alignment_TestBenchmark_Equalized\" + iscorrect;
int count = 0;
DirectoryInfo fatherInfo = new DirectoryInfo(fatherDir);
DirectoryInfo[] childrenInfos = fatherInfo.GetDirectories();
foreach (var childrenInfo in childrenInfos)
{
FileInfo[] files = childrenInfo.GetFiles("*.jpg", SearchOption.AllDirectories);
if (files.Length != 2)
{
continue;
}
else
{
Bitmap[] bits1 = new[] { new Bitmap(files[0].FullName) };
Bitmap[] bits2 = new[] { new Bitmap(files[1].FullName) };
float[] a = model.ExtractBitmapOutputs(bits1, "ip1", 0);
float[] b = model.ExtractBitmapOutputs(bits2, "ip1", 0);
float confidency = CaffeModel.CosineDistanceProb(a, b);
Write(@"C:\Users\BALTHASAR\Documents\MATLAB\FaceTest\" + iscorrect + ".txt", confidency.ToString());
}
count++;
Console.WriteLine(count);
}
}
public static RecogReply PredictionFunc(Guid id, int timeBudgetInMS, RecogRequest req)
{
byte[] imgBuf = req.Data;
byte[] imgType = System.Text.Encoding.UTF8.GetBytes("jpg");
Guid imgID = BufferCache.HashBufferAndType(imgBuf, imgType);
string imgFileName = imgID.ToString() + ".jpg";
string filename = Path.Combine(saveImageDir, imgFileName);
if (!File.Exists(filename))
{
FileTools.WriteBytesToFileConcurrent(filename, imgBuf);
}
Stopwatch timer = Stopwatch.StartNew();
CaffeModel.SetDevice(gpu);
string resultString = predictor.Predict(filename);
timer.Stop();
File.Delete(filename);
numImageRecognized++;
Console.WriteLine("Image {0}:{1}:{2}: {3}", numImageRecognized, imgFileName, timer.Elapsed, resultString);
return(VHubRecogResultHelper.FixedClassificationResult(resultString, resultString));
}
private static float[] ImageAligement(Bitmap bit, float[][] feature, int j, CaffeModel caffe5IPTs, int deviceId)
{
try
{
Image <Bgr, byte> image = new Image <Bgr, byte>(bit);
Rectangle draw = new Rectangle(Convert.ToInt32(feature[0][j * 4 + 0] * image.Width),
Convert.ToInt32(feature[0][j * 4 + 1] * image.Height),
Convert.ToInt32(feature[0][j * 4 + 2] * image.Width - feature[0][j * 4 + 0] * image.Width),
Convert.ToInt32(feature[0][j * 4 + 3] * image.Height - feature[0][j * 4 + 1] * image.Height));
//image.Draw(draw, new Bgr(Color.Aqua), image.Width/64);
//
int Margin_Height = Convert.ToInt32(draw.Height * 1.2f);
int Margin_X = Convert.ToInt32(draw.X - draw.Width * 0.1f);
int Margin_Y = Convert.ToInt32(draw.Y - draw.Height * 0.1f);
Rectangle Margindraw = new Rectangle(Margin_X, Margin_Y, Margin_Height, Margin_Height);//What's you really need
//image.Draw(Margindraw, new Bgr(Color.Crimson), image.Width / 64);
//*****************
image.ROI = Margindraw;
Image <Bgr, byte> C = new Image <Bgr, byte>(image.Bitmap);
float[] pts5 = caffe5IPTs.ExtractBitmapOutputs(new[] { C.Bitmap }, "fc2", deviceId);
for (int i = 0; i < pts5.Length; i += 2)
{
pts5[i] = (pts5[i] * Margindraw.Width - Margindraw.Width / 12.0f + draw.X - image.Width / 7.0f) / image.Width / 1.4f;
pts5[i + 1] = (pts5[i + 1] * Margindraw.Height - Margindraw.Height / 12.0f + draw.Y - image.Height / 7.0f) / image.Height / 1.4f;
}
return(pts5);
}
catch (Exception)
{
return(null);
}
}
void whatever()
{
string net1 = @"IPBBox_deploy.prototxt";
string model1 = @"IPBBox_iter_iter_60000.caffemodel";
CaffeModel IPBbox = new CaffeModel(net1, model1, false);
string net2 = @"5IPTs_deploy.prototxt";
string model2 = @"5IPTs_iter_iter_300000.caffemodel";
CaffeModel IPTs5 = new CaffeModel(net2, model2, false);
FastFace ff = new FastFace(1.1f, 3, 24);
Bitmap bmp = new Bitmap(@"D:\小咖秀\20153月296月10\20170115150220800-315.jpg");
FaceInfo info = ff.Facedetect_Multiview_Reinforce(bmp);
//Bitmap[] bits = new Bitmap[info.count];
Bitmap[] bits = new[] { bmp };
Rectangle rect = new Rectangle(0, 0, bmp.Width, bmp.Height);
int StartX = Convert.ToInt32(rect.X - rect.Width * 0.2);
int StartY = Convert.ToInt32(rect.Y - rect.Height * 0.2);
int iWidth = Convert.ToInt32(rect.Width * 1.4);
int iHeight = Convert.ToInt32(rect.Height * 1.4);
bits[0] = KiCut(bmp, StartX, StartY, iWidth, iHeight);
for (int i = 0; i < info.count; i++)
{
//int StartX = Convert.ToInt32(info.r[i].X - info.r[i].Width * 0.2);
//int StartY = Convert.ToInt32(info.r[i].Y - info.r[i].Height * 0.2);
//int iWidth = Convert.ToInt32(info.r[i].Width * 1.4);
//int iHeight = Convert.ToInt32(info.r[i].Height * 1.4);
//bits[i] = KiCut(bmp, StartX, StartY, iWidth, iHeight);
//Image<Bgr, byte> image = new Image<Bgr, byte>(bmp);
//Rectangle rect = info.r[i];
//rect.X = Convert.ToInt32(rect.X - rect.Width * 0.2);
//rect.Y = Convert.ToInt32(rect.Y - rect.Height * 0.2);
//rect.Width = Convert.ToInt32(rect.Width * 1.4);
//rect.Height = Convert.ToInt32(rect.Height * 1.4);
//image.ROI = rect;
//bits[i] = image.Bitmap;
//image.ROI = Rectangle.Empty; ;
}
if (/*info.count>0*/ true)
{
float[][] aa = IPBbox.ExtractBitmapOutputs(bits, new[] { "fc2", "fc3" }, 0);
Rectangle[] rects = new Rectangle[info.count];
Bitmap[] C = CaffeModel.Align_Step1(bits, rects, aa[0], aa[1]);
float[] bb = IPTs5.ExtractBitmapOutputs(C, "fc2", 0);
Bitmap[] F = CaffeModel.Align_Step2(bits, C, bb, rects, 128, 128);
F[0].Save(@"C:\Users\BALTHASAR\Desktop\test.jpg");
float[][] a =
IPBbox.ExtractFileOutputs(new[] { @"D:\Research\FacialLandmarks\Data\5PTS\batch1_1.jpg" },
new[] { "fc2", "fc3" }, 0);
}
}
public static bool InitializeRecognizer(VHubBackendStartParam pa)
{
var bInitialized = true;
var x = ImageCaptionInstance.Current;
if (!Object.ReferenceEquals(x, null))
{
/// <remarks>
/// To implement your own image recognizer, please obtain a connection Guid by contacting [email protected]
/// </remarks>
x.RegisterAppInfo(new Guid("843EF294-C635-42DA-9AD8-E79E82F9A357"), "0.0.0.1");
Func <Guid, int, RecogRequest, RecogReply> del = PredictionFunc;
/// <remarks>
/// Register your prediction function here.
/// </remarks>
string exeDir = Path.GetDirectoryName(Process.GetCurrentProcess().MainModule.FileName);
x.RegisterClassifierCS("#ImageCaption", Path.Combine(exeDir, "logo.jpg"), 100, del);
string wordDetectorProto = Path.Combine(rootDir, @"Model\wordDetector.proto");
string wordDetectorModel = Path.Combine(rootDir, @"Model\wordDetector.caffemodel");
string featureDetectorProto = Path.Combine(rootDir, @"Model\FeaturesDetector.proto");
string featureDetectorModel = Path.Combine(rootDir, @"Model\FeaturesDetector.caffemodel");
string thresholdFile = Path.Combine(rootDir, @"Model\thresholds.txt");
string wordLabelMapFile = Path.Combine(rootDir, @"Model\labelmap.txt");
string languageModel = Path.Combine(rootDir, @"Model\sentencesGeneration.lblmmodel");
string referenceDssmModelFile = Path.Combine(rootDir, @"Model\reference.dssmmodel");
string candidateDssmModelFile = Path.Combine(rootDir, @"Model\candidate.dssmmodel");
string trigramFile = Path.Combine(rootDir, @"Model\coco.cap.l3g");
CaffeModel.SetDevice(gpu); //This needs to be set before instantiating the net
var lmTestArguments = new LangModel.Arguments
{
NumWorkers = 8,
MaxSentenceLength = 19,
NumSentences = 500,
BeamWidth = 200,
AttributesCoverageBar = 5,
KTopWords = 100
};
predictor = new Predictor(wordDetectorProto, wordDetectorModel, featureDetectorProto, featureDetectorModel,
thresholdFile, wordLabelMapFile,
languageModel, lmTestArguments, referenceDssmModelFile, candidateDssmModelFile, trigramFile);
}
else
{
bInitialized = false;
}
return(bInitialized);
}
static void Main(string[] args)
{
//CaffeModel temp = new CaffeModel(@"C:\Users\BALTHASAR\Desktop\Feature.prototxt", @"C:\Users\BALTHASAR\Desktop\Feature.caffemodel");
CaffeModel resnet101 = new CaffeModel(@"C:\Research\FaceRecognition\model\centerloss_flip_deploy.prototxt", @"C:\Research\FaceRecognition\snapshot\centerloss34_fine_iter_20000.caffemodel", 0);
//resnet101.GetUsefulPart(@"C:\Research\FaceRecognition\snapshot\centerloss34_usefulpart.caffemodel");
//SortedList<int, float> results = lfw_fe(@"D:\Research\FaceRecognition\snapshot\GlasssixNet_train_iter_30000.caffemodel",
// "GlasssixNet.prototxt", "lfw_pairs.txt", @"D:\Alignment_lfw_Equalized\");
//float best_th;
//float ACC = lfw_acc(results, out best_th);
//lfw_error(best_th, results);
//Console.WriteLine(ACC);
getTestBenchamrkRes(resnet101, "Wrong");
getTestBenchamrkRes(resnet101, "Correct");
}
public void Init(string protoFile, string modelFile, string meanFile, string labelMapFile, string entityInfoFile, int gpu)
{
// Init caffe model
CaffeModel.SetDevice(gpu);
_caffeModel = new CaffeModel(protoFile, modelFile);
_caffeModel.SetMeanFile(meanFile);
Console.WriteLine("Succeed: Load Model File!\n");
_labelMap = File.ReadAllLines(labelMapFile)
.Select(line => line.Split('\t')[0])
.ToArray();
if (!string.IsNullOrEmpty(entityInfoFile))
{
_entityInfo = File.ReadLines(entityInfoFile)
.Select(line => line.Split('\t'))
.ToDictionary(cols => cols[0], cols => cols[1]);
}
}
static SortedList <int, float> lfw_fe(string model, string net, string pair, string imgDir)
{
SortedList <int, float> results = new SortedList <int, float>();
var fe = new CaffeModel(net, model);
StreamReader sr = new StreamReader(pair, Encoding.Default);
string line;
int label = 1;
while ((line = sr.ReadLine()) != null)
{
string[] sArray = line.Split('\t');
if (sArray.Length == 3)
{
DirectoryInfo child = new DirectoryInfo(imgDir + sArray[0]);
FileInfo[] file = child.GetFiles("*.jpg", SearchOption.AllDirectories);
float[] a = fe.ExtractFileOutputs(new[] { file[Convert.ToInt32(sArray[1]) - 1].FullName }, "eltmax_fc5", 0);
float[] b = fe.ExtractFileOutputs(new[] { file[Convert.ToInt32(sArray[2]) - 1].FullName }, "eltmax_fc5", 0);
float confidency = CaffeModel.CosineDistanceProb(a, b);
results.Add(label, confidency);
}
if (sArray.Length == 4)
{
DirectoryInfo child = new DirectoryInfo(imgDir + sArray[0]);
FileInfo[] file = child.GetFiles("*.jpg", SearchOption.AllDirectories);
float[] a = fe.ExtractFileOutputs(new[] { file[Convert.ToInt32(sArray[1]) - 1].FullName }, "eltmax_fc5", 0);
//
child = new DirectoryInfo(imgDir + sArray[2]);
file = child.GetFiles("*.jpg", SearchOption.AllDirectories);
float[] b = fe.ExtractFileOutputs(new[] { file[Convert.ToInt32(sArray[3]) - 1].FullName }, "eltmax_fc5", 0);
float confidency = CaffeModel.CosineDistanceProb(a, b);
results.Add(-1 * label, confidency);
}
Console.WriteLine(label);
label++;
}
return(results);
}
static async void XiaoKaXiu()
{
string model1 = @"IPBBox_iter_iter_60000.caffemodel";
string net1 = @"IPBBox_deploy.prototxt";
string model2 = @"5IPTs_iter_iter_300000.caffemodel";
string net2 = @"5IPTs_deploy.prototxt";
var IPBBox = new CaffeModel(net1, model1, false);
var IPTs = new CaffeModel(net2, model2, false);
var ff = new FastFace(1.1f, 3, 48);
string fatherDir = @"D:\CASIA\Cleaned_CASIA";
FaceServiceClient faceServiceClient = new FaceServiceClient(SubscriptionKey);
int count = 0;
//Random rd=new Random(45985369);
DirectoryInfo fatherInfo = new DirectoryInfo(fatherDir);
DirectoryInfo[] childrenInfos = fatherInfo.GetDirectories();
foreach (var childrenInfo in childrenInfos)
{
//DirectoryInfo[] SubInfos = childrenInfo.GetDirectories();
FileInfo[] files = childrenInfo.GetFiles("*.jpg", SearchOption.AllDirectories);
int filecounter = 0;
foreach (var file in files)
{
filecounter++;
try
{
if (count <= -1)
{
continue;
}
Bitmap bmp = new Bitmap(file.FullName);
FaceInfo info = ff.Facedetect_Frontal_Surveillance(bmp);
if (info.count != 1)
{
//Write("lost.txt", file.FullName);
continue;
}
////**************
Bitmap[] bits = new Bitmap[1];
bits[0] = bmp;
Rectangle rect = new Rectangle(0, 0, bmp.Width, bmp.Height);
int StartX = Convert.ToInt32(0 - bmp.Width * 0.2);
int StartY = Convert.ToInt32(0 - bmp.Height * 0.2);
int iWidth = Convert.ToInt32(bmp.Width * 1.4);
int iHeight = Convert.ToInt32(bmp.Height * 1.4);
bits[0] = KiCut(bmp, StartX, StartY, iWidth, iHeight);
float[][] feature = IPBBox.ExtractBitmapOutputs(bits, new[] { "fc2", "fc3" }, 0);
if (Math.Abs(feature[1][0] * 90) >= 30)
{
bits[0].Dispose();
bmp.Dispose();
continue;
}
Rectangle[] rects = new Rectangle[bits.Length];
Bitmap[] C = CaffeModel.Align_Step1(bits, rects, feature[0], feature[1]);
float[] bb = IPTs.ExtractBitmapOutputs(C, "fc2", 0);
Bitmap[] F = CaffeModel.Align_Step2(bits, C, bb, rects, 128, 128);
if (!Directory.Exists(@"C:\Data\Alignment_CASIA\" + childrenInfo.Name))
{
Directory.CreateDirectory(@"C:\Data\Alignment_CASIA\" + childrenInfo.Name);
}
Image <Gray, byte> gray = new Image <Gray, byte>(F[0]);
//Image<Gray, byte> temp = gray.Resize(32, 32, Inter.Linear);
//temp = temp.Resize(64, 64, Inter.Linear);
gray._EqualizeHist();
Image <Bgr, byte> colorimg = new Image <Bgr, byte>(new Image <Gray, byte>[] { gray, gray, gray });
colorimg.Save(@"C:\Data\Alignment_CASIA\" + childrenInfo.Name + "\\" + filecounter + ".jpg");
//temp._EqualizeHist();
//temp.Save(@"C:\Data\Alignment_Registed_Equalized\" + childrenInfo.Name + "\\" + filecounter + "_per_blur.jpg");
//Console.WriteLine(@"C:\Data\Alignment_CASIA_Equalized\" + childrenInfo.Name + "\\" + filecounter + " complete!");
////*****
colorimg.Dispose();
gray.Dispose();
F[0].Dispose();
C[0].Dispose();
bits[0].Dispose();
bmp.Dispose();
}
catch (Exception ex)
{
Console.WriteLine(ex.ToString());
continue;
throw;
}
//}
//Console.WriteLine(filecounter);
}
Console.WriteLine(count);
count++;
}
}