changed face detection model to yolo_mobilenet_backbone

app/__init__.py

@@ -9,7 +9,8 @@ from face_recognition import helper as fr_helper
 
 
 
-face_detector=fd.face_detection("face_detection/Models/v1")
+# face_detector=fd.face_detection("face_detection/Models/v1")
+face_detector=fd.face_detection("face_detection/Models/mobilenet")
 face_detector.square_preprocessing=fd.square_pad()
 # face_recognizer=fr.face_recognition("face_recognition/Models/v1")
 # face_recognizer=fr.face_recognition("face_recognition/Models/mobilenet_basic_lfw")

face_detection/Models/mobilenet/anchors.txt

@@ -0,0 +1,5 @@
+1.374126791954040527e+00 1.648838758468627930e+00
+3.987907886505126953e+00 4.713618278503417969e+00
+2.353522062301635742e+00 2.840777397155761719e+00
+6.863890647888183594e+00 8.402585983276367188e+00
+6.510864496231079102e-01 7.812500000000000000e-01

face_detection/Models/mobilenet/config.py

@@ -0,0 +1,2 @@
+p_thres=0.5
+nms_thres=0.3

face_detection/Models/mobilenet/model.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ee8452826fb51076f13878d77119bd7834a946df3fb396002c7b756cd07ba64e
+size 13201720

face_detection/Models/v1/model.h5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:90dd2a59c7acc25d967c9914a7a53362ed2229c56e3105f538e04388fce70390
-size 607092424
+oid sha256:5a50e569a9d63bd7e30fb1107b3b9164e2f9e2512d4c98fd13f04ccad15365b2
+size 202503752

face_detection/create_load_model.py

@@ -4,251 +4,38 @@ from tensorflow.keras import Model,layers
 from face_detection.config import class_names
 
 os.environ['CUDA_VISIBLE_DEVICES'] = '-1'
-# custom layer for reshaping last layer
 
 # custom layer for reshaping last layer
-
 class yolo_reshape(tf.keras.layers.Layer):
-  global num_anchors
-  def __init__(self, **kwargs):
-    super(yolo_reshape,self).__init__()
-    self.last_item=(5+len(class_names))
-  def call(self,output_layer):
-    shape = [tf.shape(output_layer)[k] for k in range(4)]
-#     print(shape)
-#     tf.print(shape)
-    return tf.reshape(output_layer,[shape[0],shape[1],shape[2],num_anchors,self.last_item])
-
-def create_model():
-    global num_anchors
-    def space_to_depth_x2(x):
-        return tf.nn.space_to_depth(x,block_size=2)
-
-    x_input=layers.Input(shape=(None,None,3))
-    x=layers.Lambda(lambda x:x/255.)(x_input)
-    x=layers.Conv2D(32,(3,3),strides=(1,1),padding='same',name='conv_1',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_1')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-    x=layers.MaxPooling2D(pool_size=(2,2))(x)
-
-    x=layers.Conv2D(64,(3,3),strides=(1,1),padding='same',name='conv_2',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_2')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-    x=layers.MaxPooling2D(pool_size=(2,2))(x)
-
-    x=layers.Conv2D(128,(3,3),strides=(1,1),padding='same',name='conv_3',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_3')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-
-    x=layers.Conv2D(64,(1,1),strides=(1,1),padding='same',name='conv_4',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_4')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-
-    x=layers.Conv2D(128,(3,3),strides=(1,1),padding='same',name='conv_5',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_5')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-    x=layers.MaxPooling2D(pool_size=(2,2))(x)
-
-    x=layers.Conv2D(256,(3,3),strides=(1,1),padding='same',name='conv_6',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_6')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-
-    x=layers.Conv2D(128,(1,1),strides=(1,1),padding='same',name='conv_7',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_7')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-
-    x=layers.Conv2D(256,(3,3),strides=(1,1),padding='same',name='conv_8',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_8')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-    x=layers.MaxPooling2D(pool_size=(2,2))(x)
-
-    x=layers.Conv2D(512,(3,3),strides=(1,1),padding='same',name='conv_9',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_9')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-
-    x=layers.Conv2D(256,(1,1),strides=(1,1),padding='same',name='conv_10',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_10')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-
-    x=layers.Conv2D(512,(3,3),strides=(1,1),padding='same',name='conv_11',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_11')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-
-    x=layers.Conv2D(256,(1,1),strides=(1,1),padding='same',name='conv_12',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_12')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-
-    x=layers.Conv2D(512,(3,3),strides=(1,1),padding='same',name='conv_13',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_13')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-
-    skip_connection = x
-
-    x=layers.MaxPooling2D(pool_size=(2,2))(x)
-
-    x=layers.Conv2D(1024,(3,3),strides=(1,1),padding='same',name='conv_14',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_14')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-
-    x=layers.Conv2D(512,(1,1),strides=(1,1),padding='same',name='conv_15',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_15')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-
-    x=layers.Conv2D(1024,(3,3),strides=(1,1),padding='same',name='conv_16',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_16')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-
-    x=layers.Conv2D(512,(1,1),strides=(1,1),padding='same',name='conv_17',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_17')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-
-    x=layers.Conv2D(1024,(3,3),strides=(1,1),padding='same',name='conv_18',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_18')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-
-    x=layers.Conv2D(1024,(3,3),strides=(1,1),padding='same',name='conv_19',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_19')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-
-    x=layers.Conv2D(1024,(3,3),strides=(1,1),padding='same',name='conv_20',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_20')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
 
-    skip_connection=layers.Conv2D(64,(1,1),strides=(1,1),padding='same',name='conv_21',use_bias=False)(skip_connection)
-    skip_connection=layers.BatchNormalization(name='norm_21')(skip_connection)
-    skip_connection=layers.LeakyReLU(alpha=0.1)(skip_connection)
-    skip_connection=layers.Lambda(space_to_depth_x2)(skip_connection) # halfs the resolution and add more depth
-
-    x=layers.concatenate([skip_connection,x])
-    x=layers.Conv2D(1024,(3,3),strides=(1,1),padding='same',name='conv_22',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_22')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-
-    x=layers.Conv2D((num_anchors*(5+len(class_names))),(1,1),strides=(1,1),padding='same',name='conv_23')(x)
-    out=yolo_reshape()(x)
-
-    model=Model(x_input,out,name='yolo_v2_model')
-    # model.summary()
-    return model
-
-def create_tiny_model():
-    global num_anchors
-
-    x_input=layers.Input(shape=(416,416,3))
-    x=layers.Lambda(lambda x:x/255.)(x_input)
-    x=layers.Conv2D(16,(3,3),strides=(1,1),padding='same',name='conv_1',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_1')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-    x=layers.MaxPooling2D(pool_size=(2,2),strides=(2,2))(x)
-
-    x=layers.Conv2D(32,(3,3),strides=(1,1),padding='same',name='conv_2',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_2')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-    x=layers.MaxPooling2D(pool_size=(2,2),strides=(2,2))(x)
-
-    x=layers.Conv2D(64,(3,3),strides=(1,1),padding='same',name='conv_3',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_3')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-    x=layers.MaxPooling2D(pool_size=(2,2),strides=(2,2))(x)
-
-    x=layers.Conv2D(128,(3,3),strides=(1,1),padding='same',name='conv_4',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_4')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-    x=layers.MaxPooling2D(pool_size=(2,2),strides=(2,2))(x)
-
-    x=layers.Conv2D(256,(3,3),strides=(1,1),padding='same',name='conv_5',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_5')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-    x=layers.MaxPooling2D(pool_size=(2,2),strides=(2,2))(x)
-
-    x=layers.Conv2D(512,(3,3),strides=(1,1),padding='same',name='conv_6',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_6')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-    # x=layers.MaxPooling2D(pool_size=(2,2),strides=(1,1))(x)
-
-    x=layers.Conv2D(1024,(3,3),strides=(1,1),padding='same',name='conv_7',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_7')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
+  def __init__(self,num_anchors,last_item, **kwargs):
+    super(yolo_reshape, self).__init__(**kwargs)    
+    self.last_item=last_item
+    self.num_anchors=num_anchors
 
+  def call(self,output_layer):
+    shape = [tf.shape(output_layer)[k] for k in range(4)]
+    return tf.reshape(output_layer,[shape[0],shape[1],shape[2],self.num_anchors,self.last_item])
+  
+  def compute_output_shape(self, input_shape):
+    return (input_shape[0],input_shape[1],input_shape[2],num_anchors,self.last_item, self.num_anchors,self.last_item)
+
+  
+  def get_config(self):
+    config = super(yolo_reshape, self).get_config()
+    config.update(
+        {
+            "last_item": self.last_item,
+            "num_anchors": self.num_anchors
+        }
+    )
+    return config
     
-    x=layers.Conv2D(1024,(3,3),strides=(1,1),padding='same',name='conv_8',use_bias=False)(x)
-    x=layers.BatchNormalization(name='norm_8')(x)
-    x=layers.LeakyReLU(alpha=0.1)(x)
-
-    x=layers.Conv2D((num_anchors*(5+len(class_names))),(1,1),strides=(1,1),padding='same',name='conv_9')(x)
-    out=yolo_reshape()(x)
-
-    model=Model(x_input,out,name='yolo_v2_tiny_model')
-    # model.summary()
-    return model
+  @classmethod
+  def from_config(cls, config):
+    return cls(**config)
+  
 
 def load_model(path):
-    # model=tf.keras.models.load_model(path,custom_objects={'yolo_dynamic_reshape':yolo_dynamic_reshape},compile=False)
-    model=create_model()
-    model.load_weights(path)
-    # model=tf.keras.models.load_model(path)
+    model=tf.keras.models.load_model(path,compile=False,custom_objects={"yolo_reshape":yolo_reshape})
     return model
-
-
-import struct
-
-def get_original_weights(model,path_to_weight = "./yolov2-voc.weights"):
-  global num_anchors
-  # path_to_weight = "./yolov2.weights"(trained on coco dataset (80 classes))
-  if "yolov2-voc.weights" in path_to_weight: offset=5;nb_conv = 23;
-  if "darknet19_448.conv.23" in path_to_weight: offset=4;nb_conv = 18;
-  if "yolov2-tiny-voc.weights" in path_to_weight: offset=4;nb_conv = 9;
-  print(offset,nb_conv)
-
-  class WeightReader:
-          def __init__(self, weight_file):
-            self.offset = offset # an offset of 5 as first 5 values are non weight values(they are weight header)(for yolov2-voc.weights)
-    #         self.all_weights = np.fromfile(weight_file, dtype='float32')
-            self.all_weights = open(weight_file,'rb')
-            weight_header=struct.unpack(f'{offset}i', self.all_weights.read(offset*4))
-            # print("weight Header(major, minor, revision, seen):",weight_header)
-                  
-          def read_bytes(self, size):
-            weights = struct.unpack('%df' % size, self.all_weights.read(size*4))
-            # print(weights)
-            # input("wait now forever")
-            return np.array(weights)
-                  
-  weight_reader = WeightReader(path_to_weight)
-  # print("all_weights = {}".format(np.fromfile(path_to_weight, dtype='float32').shape[0]-weight_reader.offset))
-
-  for i in range(1, nb_conv+1):
-    conv_layer = model.get_layer('conv_' + str(i))
-    
-    if i < nb_conv:
-      norm_layer = model.get_layer('norm_' + str(i))
-      
-      size = np.prod(norm_layer.get_weights()[0].shape)
-
-      beta  = weight_reader.read_bytes(size)
-      gamma = weight_reader.read_bytes(size)
-      mean  = weight_reader.read_bytes(size)
-      var   = weight_reader.read_bytes(size)
-
-      weights = norm_layer.set_weights([gamma, beta, mean, var])       
-            
-    if len(conv_layer.get_weights()) > 1:
-      bias   = weight_reader.read_bytes(np.prod(conv_layer.get_weights()[1].shape))
-      kernel = weight_reader.read_bytes(np.prod(conv_layer.get_weights()[0].shape))
-      kernel = kernel.reshape(list(reversed(conv_layer.get_weights()[0].shape)))
-      kernel = kernel.transpose([2,3,1,0])
-      # print(kernel.shape)
-      kernel=kernel.reshape([*kernel.shape[:-1],num_anchors,-1]) # reshape to this format so we change change position of p idx
-      idx=4 # in darknet each object was encoded as [x,y,w,h,p,c] but we use [p,x,y,w,h,c]
-      kernel=np.concatenate([kernel[...,idx:idx+1],kernel[...,:idx],kernel[...,idx+1:]],axis=-1)  # setting p to idx 0
-      # print(kernel.shape)
-      kernel=kernel.reshape([*kernel.shape[:-2],-1])
-      # print(kernel.shape)
-      conv_layer.set_weights([kernel, bias])
-    else:
-      kernel = weight_reader.read_bytes(np.prod(conv_layer.get_weights()[0].shape))
-      kernel = kernel.reshape(list(reversed(conv_layer.get_weights()[0].shape)))
-      kernel = kernel.transpose([2,3,1,0])
-      conv_layer.set_weights([kernel])
-
-  return model

test.py

@@ -12,25 +12,23 @@ def new_distance(vectors):
 path="face_recognition/Models/mobilenet_basic_lfw/mobilenet_basic_lfw_model.h5"
 # path="face_recognition/Models/keras_mobilenet_emore_adamw/keras_mobilenet_emore_adamw.h5"
 
-model2=tf.keras.models.load_model(path,compile=False)
+model=tf.keras.models.load_model(path,compile=False)
 
 x_input=tf.keras.layers.Input(shape=(112,112,3))
 x_preprocess=tf.keras.layers.Lambda(lambda x:(x - 127.5) * 0.0078125)(x_input)
-x=model2(x_preprocess,training=False)
+x=model(x_preprocess,training=False)
 x=tf.keras.layers.Lambda(lambda x:tf.math.l2_normalize(x, axis=-1))(x)
 model2=tf.keras.Model(x_input,x)
 # model2.summary()
+# model.save(path.rsplit("/",1)[0]+"/model.h5")
+# print("model save to :\t",path.rsplit("/",1)[0]+"/model.h5")
 
 
-# cv2.imread("")
-# for layer in model2.layers:
-#     if isinstance(layer,tf.keras.layers.Activation):
-#         print(layer.activation)
 
 img1=cv2.resize(cv2.imread("C:/Users/Home/Desktop/college stuff/vasu_dataset/all/priyanshi/7.jpg")[:,:,::-1],[112,112])[None]
 img2=cv2.resize(cv2.imread("C:/Users/Home/Desktop/college stuff/vasu_dataset/all/shivansh/12.jpg")[:,:,::-1],[112,112])[None]
-v1=(model2.predict(img1,verbose=0))
-v2=(model2.predict(img2,verbose=0))
+v1=(model.predict(img1,verbose=0))
+v2=(model.predict(img2,verbose=0))
 
 
 print("euclidean_distance:",euclidean_distance([v1,v2]))
@@ -38,9 +36,4 @@ print("new_distance:",new_distance([v1,v2]))
 if(new_distance([v1,v2])[0]>0.2830035090446472):
     print("matching")
 else:
-    print("not matching")
-
-
-
-model2.save(path.rsplit("/",1)[0]+"/model.h5")
-print("model save to :\t",path.rsplit("/",1)[0]+"/model.h5")
+    print("not matching")