Add files via upload

Author: sumitgoyal-ai

FaceDetection_vgg_100.h5

@@ -0,0 +1,79 @@
+# -*- coding: utf-8 -*-
+"""
+Created on Mon Sep 27 16:56:06 2021
+
+@author: sumitg
+"""
+
+import cv2
+import numpy as np
+from keras.backend.tensorflow_backend import set_session
+import tensorflow as tf
+import matplotlib.pyplot as plt
+from keras import callbacks
+config = tf.ConfigProto()
+config.gpu_options.allow_growth = True
+#config.gpu_options.per_process_gpu_memory_fraction = 0.3
+set_session(tf.Session(config=config))
+
+sift = cv2.SIFT_create()
+bf = cv2.BFMatcher()
+
+key_features = []
+for i in range(100):
+  
+    img = cv2.imread(r"\\192.168.4.37\Sumit\Face\img/"+str(i)+".bmp")
+    
+    kp1, des1 = sift.detectAndCompute(img,None)
+    if des1 is None:
+        print(i)
+    bConsider = False
+    avg = []
+    maxx = []
+    for des in key_features:
+        scores = []
+        for d in des:
+            des2,kp2,img2 = d
+            matches = bf.knnMatch(des1,des2, k=2)
+            good = []
+            for m,n in matches:
+                if m.distance < 0.8*n.distance:
+                    good.append([m])
+            if len(matches)==0:
+                scores.append(0.0)
+            else:
+                scores.append(len(good)/len(matches))
+            
+# =============================================================================
+#             img3 = cv2.drawMatchesKnn(img,kp1,img2,kp2,good,None,flags = 2)
+# 
+#             plt.imshow(img3),plt.show()
+# =============================================================================
+        avg_score = np.average(scores)
+        max_score = np.max(scores)
+        
+        
+        avg.append(avg_score)
+        maxx.append(max_score)
+    if len(avg)>0:
+        armax_avg = np.argmax(maxx)
+        if maxx[armax_avg]>=.20:
+            bConsider = True
+            key_features[armax_avg].append((des1,kp1,img)) 
+    if bConsider == False:
+        key_features.append([(des1,kp1,img)]) 
+        print(len(key_features),i)
+    
+for cat in range(len(key_features)):
+    print(cat)
+    rows = (len(key_features[cat])/3)+1
+    fig = plt.figure(figsize=(rows, 3))
+               
+    for i in range(len(key_features[cat])):
+        fig.add_subplot(rows, 3, i+1)
+        plt.imshow(key_features[cat][i][2])
+# =============================================================================
+#         plt.imshow(key_features[cat][i][2])
+#         plt.show()
+# =============================================================================
+        

FaceDetection_vgg_bf.h5

@@ -0,0 +1,250 @@
+# -*- coding: utf-8 -*-
+"""
+Created on Fri Sep 24 09:45:38 2021
+
+@author: sumitgoyal
+"""
+import face_recognition
+import os
+import cv2
+import random
+import scipy.io
+import numpy as np
+import matplotlib.pyplot as plt
+from numpy import pi, exp, sqrt
+from keras.applications.resnet50 import ResNet50
+from keras.applications.resnet50 import preprocess_input
+from keras.applications.vgg16 import VGG16
+from keras.applications.vgg16 import preprocess_input
+from keras.models import Model
+from keras.callbacks import ModelCheckpoint
+from keras.backend.tensorflow_backend import set_session
+import tensorflow as tf
+from keras import callbacks
+config = tf.ConfigProto()
+config.gpu_options.allow_growth = True
+#config.gpu_options.per_process_gpu_memory_fraction = 0.3
+set_session(tf.Session(config=config))
+
+
+
+from keras.layers import MaxPool2D,BatchNormalization,Conv2D,Activation,UpSampling2D,Concatenate 
+s, k = 500, 500 
+probs = [exp(-z*z/(2*s*s))/sqrt(2*pi*s*s) for z in range(-k,k+1)] 
+kernel = np.outer(probs, probs)
+kernel = kernel-np.min(kernel)
+maxc = np.max(kernel)
+kernel = kernel/maxc
+kernel[kernel<.50]=0
+
+base_dir = 'faces/'
+dims = (224,344)
+validation_split = .10
+thres = .5
+if os.path.exists(base_dir)==False:
+    print("Face Dataset is not available in current directory")
+    raise SystemExit
+images_path = []
+for dirpath, dname, filename in os.walk(base_dir):
+    for fname in filename:
+        if fname.endswith(".jpg"):
+            images_path.append(os.path.join(dirpath, fname))
+mat = scipy.io.loadmat(os.path.join(base_dir, 'ImageData.mat'))
+
+data = []
+
+for i,path in enumerate(images_path):
+    img = cv2.imread(path)
+    mask = np.zeros((img.shape[0],img.shape[1]),dtype='float32')
+    label = mat['SubDir_Data'][:,i]
+    xmin = int(np.min([label[0],label[2],label[4],label[6]]))
+    xmax = int(np.max([label[0],label[2],label[4],label[6]]))
+    ymin = int(np.min([label[1],label[3],label[5],label[7]]))
+    ymax = int(np.max([label[1],label[3],label[5],label[7]]))
+    mask[ymin:ymax,xmin:xmax] = cv2.resize(kernel,(xmax-xmin,ymax-ymin))
+    img = cv2.resize(img,dims)
+    mask = cv2.resize(mask,dims)
+    
+    data.append([img,mask])
+
+    
+random.seed(0)
+random.shuffle(data)
+
+nTotal = len(data)
+nValidation = int(validation_split*nTotal)
+nTraining = nTotal-nValidation
+train_X = []
+train_Y = []
+for dat in data[:nTraining]:
+    train_X.append(dat[0])
+    train_Y.append(dat[1])
+Val_X = []
+Val_Y = []
+for dat in data[nTraining:]:
+    Val_X.append(dat[0])
+    Val_Y.append(dat[1])
+    
+train_X1 = preprocess_input(np.array(train_X).astype('float32'))
+Val_X1 = preprocess_input(np.array(Val_X).astype('float32'))
+
+train_Y1 = np.expand_dims(np.array(train_Y),axis=-1)
+Val_Y1 = np.expand_dims(np.array(Val_Y),axis=-1)
+
+def getModelBaseResnet():
+
+    base_model = ResNet50(include_top=False)  
+     
+    x = UpSampling2D()(base_model.layers[44].output)
+    x = Concatenate()([x,base_model.layers[38].output])
+    
+    x = Conv2D(128,(1,1),padding='same')(x)
+    x = BatchNormalization()(x)
+    x = Activation('relu')(x)
+    
+    x = Conv2D(64,(3,3),padding='same')(x)
+    x = BatchNormalization()(x)
+    x = Activation('relu')(x)
+    
+    x = UpSampling2D()(x)
+    x = Concatenate()([x,base_model.layers[4].output])
+    
+    x = Conv2D(64,(1,1),padding='same')(x)
+    x = BatchNormalization()(x)
+    x = Activation('relu')(x)
+    
+    x = Conv2D(32,(3,3),padding='same')(x)
+    x = BatchNormalization()(x)
+    x = Activation('relu')(x)
+    
+    x = UpSampling2D()(x)
+    
+    x = Conv2D(16,(3,3),padding='same')(x)
+    x = BatchNormalization()(x)
+    x = Activation('relu')(x)
+    
+    x = Conv2D(1,(3,3),padding='same',activation='sigmoid')(x)
+    
+    
+    model = Model(base_model.layers[0].output,x)
+    for i in range(45):
+        model.layers[i].trainable = True    
+    model.summary() 
+    return model
+def getModelBaseVgg():
+
+    base_model = VGG16(include_top=False)  
+     
+    x = UpSampling2D()(base_model.layers[11].output)
+    x = Concatenate()([x,base_model.layers[6].output])
+    
+    x = Conv2D(128,(1,1),padding='same')(x)
+    x = BatchNormalization()(x)
+    x = Activation('relu')(x)
+    
+    x = Conv2D(64,(3,3),padding='same')(x)
+    x = BatchNormalization()(x)
+    x = Activation('relu')(x)
+    
+    x = UpSampling2D()(x)
+    x = Concatenate()([x,base_model.layers[3].output])
+    
+    x = Conv2D(64,(1,1),padding='same')(x)
+    x = BatchNormalization()(x)
+    x = Activation('relu')(x)
+    
+    x = Conv2D(32,(3,3),padding='same')(x)
+    x = BatchNormalization()(x)
+    x = Activation('relu')(x)
+    
+    x = UpSampling2D()(x)
+    
+    x = Conv2D(16,(3,3),padding='same')(x)
+    x = BatchNormalization()(x)
+    x = Activation('relu')(x)
+    
+    x = Conv2D(1,(3,3),padding='same',activation='sigmoid')(x)
+    
+    
+    model = Model(base_model.layers[0].output,x)
+    for i in range(12):
+        model.layers[i].trainable = True
+    model.summary() 
+    return model
+model = getModelBaseResnet()
+#model = getModelBaseVgg()
+
+
+model.load_weights("FaceDetection_resnet_100.h5")
+
+
+score_train = model.predict(train_X1)
+score_train[score_train<thres] = 0
+score_train[score_train>=thres] = 1
+
+
+sift = cv2.SIFT()
+bf = cv2.BFMatcher()
+
+key_features = []
+
+for i in range(nTraining):
+    nLabels, labels, stats, centroids = cv2.connectedComponentsWithStats(score_train[i].astype(np.uint8), connectivity=4)
+    for k in range(1,nLabels):
+        size = stats[k, cv2.CC_STAT_AREA]        
+        if size>100:
+            x, y = stats[k, cv2.CC_STAT_LEFT], stats[k, cv2.CC_STAT_TOP]
+            w, h = stats[k, cv2.CC_STAT_WIDTH], stats[k, cv2.CC_STAT_HEIGHT]
+            img  = cv2.cvtColor(train_X[i][y:y+h,x:x+w],cv2.COLOR_BGR2GRAY)
+            
+            kp1, des1 = sift.detectAndCompute(img,None)
+            if des1 is None:
+                print(i)
+            bConsider = False
+            avg = []
+            maxx = []
+            for des in key_features:
+                scores = []
+                for d in des:
+                    des2,kp2,img2 = d
+                    matches = bf.knnMatch(des1,des2, k=2)
+                    good = []
+                    for m,n in matches:
+                        if m.distance < 0.8*n.distance:
+                            good.append([m])
+                    if len(matches)==0:
+                        scores.append(0.0)
+                    else:
+                        scores.append(len(good)/len(matches))
+                    
+
+                avg_score = np.average(scores)
+                max_score = np.max(scores)
+                
+                
+                avg.append(avg_score)
+                maxx.append(max_score)
+            if len(avg)>0:
+                armax_avg = np.argmax(maxx)
+                if maxx[armax_avg]>=.20:
+                    bConsider = True
+                    key_features[armax_avg].append((des1,kp1,img)) 
+            if bConsider == False:
+                key_features.append([(des1,kp1,img)]) 
+                print(len(key_features),i)
+    
+for cat in range(len(key_features)):
+    print(cat)
+    rows = (len(key_features[cat])/3)+1
+    fig = plt.figure(figsize=(rows, 3))
+               
+    for i in range(len(key_features[cat])):
+        fig.add_subplot(rows, 3, i+1)
+        plt.imshow(key_features[cat][i][2])
+# =============================================================================
+#         plt.imshow(key_features[cat][i][2])
+#         plt.show()
+# =============================================================================
+        
+
+