new

CascadeClassifier.py

@@ -38,17 +38,17 @@ def detecvFaceImgs(self,img): #return one
 
     def getDetectImg(self,image):
         faces = self.detecvFace(image)
-        print("Found {0} faces!".format(len(faces)))
+        #print("Found {0} faces!".format(len(faces)))
         newImg = image.copy() 
         for (x, y, w, h) in faces:
-            print(x, y, w, h)
+            #print(x, y, w, h)
             cv2.rectangle(newImg, (x, y), (x+w, y+h), (0, 255, 0), 2)
 
         return newImg
 
     def showDetectImg(self,image):
         faces = self.detecvFace(image)
-        print("Found {0} faces!".format(len(faces)))
+        #print("Found {0} faces!".format(len(faces)))
 
         # Draw a rectangle around the faces
         for (x, y, w, h) in faces:

README.md

@@ -3,4 +3,11 @@
 ![Tensorflow Version](https://img.shields.io/badge/Tensorflow-V2.2.0-brightgreen)
 
 <br/>
-Facial key-points detection by using CNN model.
+Facial key-points detection by using CNN model. Dataset：FG-Net
+
+#### Keypoints regresiion
+<br/>
+<img src="images/Figure_3.png" width="320" height="240">
+<img src="images/Figure_4.png" width="320" height="240">
+<img src="images/Figure_5.png" width="320" height="240">
+<img src="images/Figure_2.png" width="320" height="240">

commonModule/ImageBase.py

@@ -30,6 +30,10 @@ def getImagChannel(img):
         return 3
     return 1  #only one channel
 
+def resizeImg(img,NewW,NewH):
+    rimg = cv2.resize(img, (NewW,NewH), interpolation=cv2.INTER_CUBIC) #INTER_CUBIC INTER_NEAREST INTER_LINEAR INTER_AREA
+    return rimg
+
 def showimage(img,str='image',autoSize=False):
     flag = cv2.WINDOW_NORMAL
     if autoSize:

commonModule/imagePlot.py

@@ -2,16 +2,57 @@
 #Steven 11/03/2020 image plot modoule
 
 import matplotlib.pyplot as plt
+import matplotlib.gridspec as gridspec
 from commonModule.common import getRowAndColumn
 
-def plotImagList(imgList,nameList,gray=False,showticks=True):
+def plotImagList2(imgList,nameList,gray=False,showTitle=True,showticks=True):
+    nImg = len(imgList)
+    nRow,nColumn = getRowAndColumn(nImg)
+
+    #f, axarr = plt.subplots(nRow, nColumn,gridspec_kw = {'wspace':0, 'hspace':0}) #
+    f, axarr = plt.subplots(nRow, nColumn, constrained_layout=True) #
+
+    gs1 = gridspec.GridSpec(nRow, nColumn)
+    gs1.update(wspace=0, hspace=0)
+
+    print(len(f.axes),nRow,nColumn)
+
+    for n in range(nImg):
+        ax = plt.subplot(gs1[n])
+        if 1:
+            img = imgList[n]
+
+            if showTitle:
+                ax.title.set_text(nameList[n])
+            if gray:
+                ax.imshow(img,cmap="gray")
+            else:
+                ax.imshow(img)
+
+            if not showticks:
+                ax.set_yticks([])
+                ax.set_xticks([])
+
+            ax.margins(0, 0) 
+            #ax.xaxis.set_major_locator(plt.NullLocator())
+            #ax.yaxis.set_major_locator(plt.NullLocator())
+
+
+    #plt.grid(True)
+    plt.tight_layout(pad=0)
+    plt.subplots_adjust(wspace=0, hspace=0)
+    #plt.subplots_adjust(top = 1, bottom = 0, right = 1, left = 0, hspace = 0, wspace = 0)
+    plt.show()
+
+def plotImagList(imgList,nameList,gray=False,showTitle=True,showticks=True):
     nImg = len(imgList)
     nRow,nColumn = getRowAndColumn(nImg)
 
     for n in range(nImg):
         img = imgList[n]
         ax = plt.subplot(nRow, nColumn, n + 1)
-        ax.title.set_text(nameList[n])
+        if showTitle:
+            ax.title.set_text(nameList[n])
         if gray:
             plt.imshow(img,cmap="gray")
         else:
@@ -20,8 +61,14 @@ def plotImagList(imgList,nameList,gray=False,showticks=True):
         if not showticks:
             ax.set_yticks([])
             ax.set_xticks([])
+
+        #ax.margins(0, 0) 
+        #ax.xaxis.set_major_locator(plt.NullLocator())
+        #ax.yaxis.set_major_locator(plt.NullLocator())
     #plt.grid(True)
     plt.tight_layout()
+    #plt.subplots_adjust(wspace=0, hspace=0)
+    #plt.subplots_adjust(top = 1, bottom = 0, right = 1, left = 0, hspace = 0, wspace = 0)
     plt.show()
 
 def main():

faceIdentification.py

@@ -16,6 +16,9 @@ def resizeImg(img,NewW,NewH):
     #return cv2.resize(img, (int(h*ratio), int(w*ratio)), interpolation=cv2.INTER_CUBIC) #INTER_LANCZOS4
     return cv2.resize(img, (NewW,NewH), interpolation=cv2.INTER_CUBIC) #INTER_CUBIC INTER_NEAREST INTER_LINEAR INTER_AREA
 
+def CascadeDetect(cascPath=r'./res/haarcascade_frontalface_default.xml'):
+    return CascadeClassifier(cascPath)
+
 def main():
     cv2.useOptimized()
     file =  r'./res/obama.jpg'#r'./res/Lenna.png' #
@@ -25,11 +28,10 @@ def main():
     if len(sys.argv)>1:
         file = sys.argv[1]
 
-    cascPath=r'./res/haarcascade_frontalface_default.xml'
 
     img = loadImg(file,mode=cv2.IMREAD_COLOR) # IMREAD_GRAYSCALE IMREAD_COLOR
 
-    faceROI = CascadeClassifier(cascPath)
+    faceROI = CascadeDetect()
 
     faceR=faceROI.getDetectImg(img)
     face = faceROI.detecvFaceImgOne(img)
@@ -43,9 +45,9 @@ def main():
 
     plotImagList(ls,names)
 
-    faceGray = resizeImg(faceGray,newW,newH)
+    #faceGray = resizeImg(faceGray,newW,newH)
     #img.writeImg(face,'./res/myface_.png')
-    writeImg(faceGray,'./res/myface_gray.png')
+    #writeImg(faceGray,'./res/myface_gray.png')
 
 if __name__=="__main__":
     main()

genLabel.py

@@ -114,8 +114,10 @@ def resizeImg(img,NewW,NewH,pts):
     for i in pts:
         x,y = i[0],i[1]
         #newPts.append((round(x*NewW/w,4),round(y*NewH/h,4)))
-        #newPts.append((x*NewW/w, y*NewH/h)) #use location  coordinates
-        newPts.append((x/w, y/h))  #use location/size ratio
+        if 0:
+            newPts.append((x*NewW/w, y*NewH/h)) #use location  coordinates
+        else:
+            newPts.append((x/w, y/h))  #use location/size ratio
     return rimg,newPts
 
 newW=364

makeDB.py

@@ -2,10 +2,13 @@
 import cv2
 import numpy as np 
 import pandas as pd
-from genLabel import loadImg,getImgHW,getFileName,getLabelFileLabels,pathsFiles
+from genLabel import loadImg,getImgHW,getFileName,getLabelFileLabels,pathsFiles,writeAnotationFile
 
 def distanceAB(a,b):
+    #print('a.shape=',a.shape)
+    #print('b.shape=',b.shape)
     return np.sqrt(np.sum((a-b)**2))
+    #return np.sqrt(np.sum(np.abs(a-b)))
 
 def calculateRatio(pts,id1,id2,id3,id4):
     assert(id1<len(pts))
@@ -14,12 +17,24 @@ def calculateRatio(pts,id1,id2,id3,id4):
     assert(id4<len(pts))
     return distanceAB(pts[id1], pts[id2])/distanceAB(pts[id3],pts[id4])
 
-def calculateFeature(pts,H,W):
+def UpdatePtsToLocation(pts,H,W):
     for i in range(len(pts)):
         #print(pts[i][0],pts[i][1])
         pts[i][0] = pts[i][0]*W
         pts[i][1] = pts[i][1]*H
+    return pts
+
+def calculateFeature(pts,H,W,file=r'test.pts'):
+    pts = UpdatePtsToLocation(pts,H,W)
+
+    if 0:
+        for i in range(len(pts)):
+            #print(pts[i][0],pts[i][1])
+            pts[i][0] = pts[i][0]*W
+            pts[i][1] = pts[i][1]*H
     #print(pts)
+
+    writeAnotationFile(file,pts)
     pts = np.array(pts)
     #print(pts.shape)
 
@@ -43,22 +58,23 @@ def calculateFeature(pts,H,W):
     F_EyeFissure  = calculateRatio(pts,28,30,27,29)
     F_VermilionHeight = calculateRatio(pts,51,66,66,57)
     F_MouthFaceWidth = calculateRatio(pts,48,54,0,14)
-    F_Noise1 = calculateRatio(pts,39,46,39,41)
-    F_Noise2 = calculateRatio(pts,39,40,39,41)
-    F_Noise3 = calculateRatio(pts,39,38,39,41)
-    F_Noise4 = calculateRatio(pts,39,38,67,41)
-    F_Noise5 = calculateRatio(pts,39,41,67,41)
+    # F_Noise1 = calculateRatio(pts,39,46,39,41)
+    # F_Noise2 = calculateRatio(pts,39,40,39,41)
+    # F_Noise3 = calculateRatio(pts,39,38,39,41)
+    # F_Noise4 = calculateRatio(pts,39,38,67,41)
+    # F_Noise5 = calculateRatio(pts,39,41,67,41)
     #print('name:',name,'Features=',F_Facial_Index,F_Mandibular_Index,F_Intercanthal,F_OrbitalWidth,F_EyeFissure,F_VermilionHeight,F_MouthFaceWidth,F_Noise1,F_Noise2,F_Noise3,F_Noise4,F_Noise5)
 
-    return [F_Facial_Index,F_Mandibular_Index,F_Intercanthal,F_OrbitalWidth,F_EyeFissure,F_VermilionHeight,F_MouthFaceWidth,F_Noise1,F_Noise2,F_Noise3,F_Noise4,F_Noise5]
+    #return [F_Facial_Index,F_Mandibular_Index,F_Intercanthal,F_OrbitalWidth,F_EyeFissure,F_VermilionHeight,F_MouthFaceWidth,F_Noise1,F_Noise2,F_Noise3,F_Noise4,F_Noise5]
+    return [F_Facial_Index,F_Mandibular_Index,F_Intercanthal,F_OrbitalWidth,F_EyeFissure,F_VermilionHeight,F_MouthFaceWidth]
 
-    data = {'F_Facial_Index':F_Facial_Index, 'F_Mandibular_Index':F_Mandibular_Index,
-            'F_Intercanthal':F_Intercanthal,'F_OrbitalWidth':F_OrbitalWidth,'F_EyeFissure':F_EyeFissure,
-            'F_VermilionHeight':F_VermilionHeight,'F_MouthFaceWidth':F_MouthFaceWidth,'F_Noise1':F_Noise1,
-            'F_Noise2':F_Noise2,'F_Noise3':F_Noise3,'F_Noise4':F_Noise4,'F_Noise5':F_Noise5}
-    #df = pd.DataFrame(data=data,index=[0])
-    #print(df)
-    return data
+    # data = {'F_Facial_Index':F_Facial_Index, 'F_Mandibular_Index':F_Mandibular_Index,
+    #         'F_Intercanthal':F_Intercanthal,'F_OrbitalWidth':F_OrbitalWidth,'F_EyeFissure':F_EyeFissure,
+    #         'F_VermilionHeight':F_VermilionHeight,'F_MouthFaceWidth':F_MouthFaceWidth,'F_Noise1':F_Noise1,
+    #         'F_Noise2':F_Noise2,'F_Noise3':F_Noise3,'F_Noise4':F_Noise4,'F_Noise5':F_Noise5}
+    # #df = pd.DataFrame(data=data,index=[0])
+    # #print(df)
+    # return data
 
 dbFile = r'.\db\facial.csv'
 def makeDb():
@@ -67,11 +83,10 @@ def makeDb():
     imgPath = base + r'\images'
     LabelPath = base + r'\labels'
 
+    imgPath = r'E:\opencv\project\facialRecognition\db\recognitionDb'
     df = pd.DataFrame()
-    columns = ['F_Facial_Index', 'F_Mandibular_Index', 'F_Intercanthal',
-       'F_OrbitalWidth', 'F_EyeFissure', 'F_VermilionHeight',
-       'F_MouthFaceWidth', 'F_Noise1', 'F_Noise2', 'F_Noise3', 'F_Noise4',
-       'F_Noise5']
+    #columns = ['F_Facial_Index', 'F_Mandibular_Index', 'F_Intercanthal','F_OrbitalWidth', 'F_EyeFissure', 'F_VermilionHeight','F_MouthFaceWidth', 'F_Noise1', 'F_Noise2', 'F_Noise3', 'F_Noise4','F_Noise5']
+    columns = ['F_Facial_Index', 'F_Mandibular_Index', 'F_Intercanthal','F_OrbitalWidth', 'F_EyeFissure', 'F_VermilionHeight','F_MouthFaceWidth']
 
     for i in pathsFiles(imgPath,'jpg'):
         #print(i)
@@ -83,10 +98,16 @@ def makeDb():
         label = LabelPath + '\\' + fileName + '.pts'
         pts = getLabelFileLabels(label)
 
-        print(fileName,'label=', label,'pts=', len(pts))
-        data = calculateFeature(pts,H,W)
-        data = np.array(data).reshape(-1,len(data))
-        line = pd.DataFrame(data,columns=columns)
+        pts = np.reshape(pts,(68,2))
+        print(fileName,'label=', label,'pts=', len(pts),'H,W=',H,W)
+        if 1:
+            data = calculateFeature(pts,H,W)
+            data = np.array(data).reshape(-1,len(data))
+            line = pd.DataFrame(data,columns=columns)
+        else:
+            pts = pts.reshape(1,136)
+            line = pd.DataFrame(pts)
+
         line.insert(0, "Id", fileName, True)
 
         df = df.append(line)

plotLoss.py

@@ -0,0 +1,78 @@
+#Steven 24/04/2020 
+import argparse 
+import sys
+import matplotlib.pyplot as plt
+
+#----------------------------------------------
+#usgae: python plotloss.py .\facialRecognition.log
+#----------------------------------------------
+
+def getLoss(log_file,startIter=0,stopIter=None):
+    numbers = {'1','2','3','4','5','6','7','8','9'}
+    with open(log_file, 'r') as f:
+        lines  = [line.rstrip("\n") for line in f.readlines()]
+
+        iters = []
+        loss = []
+        val_loss=[]
+        for line in lines:
+            trainIterRes = line.split(' ')
+            #print(line)
+            epoch = 0
+            if trainIterRes[0] == 'Epoch' and trainIterRes[1][-1:]!=':':
+                str = trainIterRes[1]
+                epoch = int(str[:str.find('/')])
+                #print(trainIterRes[1],epoch)
+                if(epoch<startIter):
+                    continue       
+                if stopIter and  epoch > stopIter:
+                    break
+
+                iters.append(epoch)
+
+            if trainIterRes[0] == '9/9' and trainIterRes[3] != 'ETA:':
+                print(line)
+                print(trainIterRes[7],trainIterRes[10])
+
+                loss.append(float(trainIterRes[7]))
+                val_loss.append(float(trainIterRes[10]))
+
+    return iters,loss,val_loss
+
+def plotLoss(ax,iters,loss,label='',name='Training loss'):
+    #ax.set_title(name)
+    ax.plot(iters,loss,label=label)
+    ax.set_xlabel('Epoch')
+    ax.set_ylabel('Loss')
+    ax.legend()
+
+def main(argv):
+    parser = argparse.ArgumentParser()
+    parser.add_argument('-l','--list', nargs='+', help='path to log file', required=True)
+    parser.add_argument('-s', '--start', help = 'startIter')
+    parser.add_argument('-t', '--stop', help = 'stopIter')
+
+    args = parser.parse_args()
+    startIter = 0
+    stopIter = None
+    if args.start:
+        startIter = int(args.start)
+    if args.stop:
+        stopIter = int(args.stop)
+
+    print(args.list,startIter,stopIter)
+
+    ax = plt.subplot(1,1,1)
+    file = args.list[0]
+    iters,loss,val_loss = getLoss(file,startIter,stopIter)
+
+    plotLoss(ax,iters,loss,label='On train set')
+    plotLoss(ax,iters,val_loss,label='On validation set')
+    #plt.ylim(0, 4)
+    plt.yscale("log")
+    plt.legend()
+    plt.show()
+
+if __name__ == "__main__":
+    main(sys.argv)
+

predictKeyPoints.py

@@ -16,7 +16,7 @@ def argCmdParse():
 
     return parser.parse_args()
 
-def preditImg(img, modelName = r'./weights/trainFacialRecognition.h5'):
+def preditImg(img, modelName = r'./weights/trainFacialRecognition.h5'): # #r'./weightLoc/trainFacialRecognition.h5'
     model = ks.models.load_model(modelName)
     print(img.shape)
     x = img[:,:,0]
@@ -25,6 +25,7 @@ def preditImg(img, modelName = r'./weights/trainFacialRecognition.h5'):
     print(x.shape)
     pts = model.predict(x)
     pts = pts.reshape((68,2))
+    print('preditImg pts.shape=',type(pts),pts.shape)
     return pts
 
 def main():