issue 131 normalization

deepface/DeepFace.py

@@ -66,7 +66,7 @@ def build_model(model_name):
 
 	return model_obj[model_name]
 
-def verify(img1_path, img2_path = '', model_name = 'VGG-Face', distance_metric = 'cosine', model = None, enforce_detection = True, detector_backend = 'opencv', align = True, prog_bar = True):
+def verify(img1_path, img2_path = '', model_name = 'VGG-Face', distance_metric = 'cosine', model = None, enforce_detection = True, detector_backend = 'opencv', align = True, prog_bar = True, normalization = 'base'):
 
 	"""
 	This function verifies an image pair is same person or different persons.
@@ -161,12 +161,16 @@ def verify(img1_path, img2_path = '', model_name = 'VGG-Face', distance_metric =
 				img1_representation = represent(img_path = img1_path
 						, model_name = model_name, model = custom_model
 						, enforce_detection = enforce_detection, detector_backend = detector_backend
-						, align = align)
+						, align = align
+						, normalization = normalization
+						)
 
 				img2_representation = represent(img_path = img2_path
 						, model_name = model_name, model = custom_model
 						, enforce_detection = enforce_detection, detector_backend = detector_backend
-						, align = align)
+						, align = align
+						, normalization = normalization
+						)
 
 				#----------------------
 				#find distances between embeddings
@@ -458,7 +462,7 @@ def analyze(img_path, actions = ['emotion', 'age', 'gender', 'race'] , models =
 
 		return resp_obj
 
-def find(img_path, db_path, model_name ='VGG-Face', distance_metric = 'cosine', model = None, enforce_detection = True, detector_backend = 'opencv', align = True, prog_bar = True):
+def find(img_path, db_path, model_name ='VGG-Face', distance_metric = 'cosine', model = None, enforce_detection = True, detector_backend = 'opencv', align = True, prog_bar = True, normalization = 'base'):
 
 	"""
 	This function applies verification several times and find an identity in a database
@@ -571,7 +575,9 @@ def find(img_path, db_path, model_name ='VGG-Face', distance_metric = 'cosine',
 					representation = represent(img_path = employee
 						, model_name = model_name, model = custom_model
 						, enforce_detection = enforce_detection, detector_backend = detector_backend
-						, align = align)
+						, align = align
+						, normalization = normalization
+						)
 
 					instance.append(representation)
 
@@ -613,7 +619,9 @@ def find(img_path, db_path, model_name ='VGG-Face', distance_metric = 'cosine',
 				target_representation = represent(img_path = img_path
 					, model_name = model_name, model = custom_model
 					, enforce_detection = enforce_detection, detector_backend = detector_backend
-					, align = align)
+					, align = align
+					, normalization = normalization
+					)
 
 				for k in metric_names:
 					distances = []
@@ -704,7 +712,7 @@ def find(img_path, db_path, model_name ='VGG-Face', distance_metric = 'cosine',
 
 	return None
 
-def represent(img_path, model_name = 'VGG-Face', model = None, enforce_detection = True, detector_backend = 'opencv', align = True):
+def represent(img_path, model_name = 'VGG-Face', model = None, enforce_detection = True, detector_backend = 'opencv', align = True, normalization = 'base'):
 
 	"""
 	This function represents facial images as vectors.
@@ -722,6 +730,8 @@ def represent(img_path, model_name = 'VGG-Face', model = None, enforce_detection
 
 		detector_backend (string): set face detector backend as retinaface, mtcnn, opencv, ssd or dlib
 
+		normalization (string): normalize the input image before feeding to model
+
 	Returns:
 		Represent function returns a multidimensional vector. The number of dimensions is changing based on the reference model. E.g. FaceNet returns 128 dimensional vector; VGG-Face returns 2622 dimensional vector.
 	"""
@@ -741,6 +751,13 @@ def represent(img_path, model_name = 'VGG-Face', model = None, enforce_detection
 		, detector_backend = detector_backend
 		, align = align)
 
+	#---------------------------------
+	#custom normalization
+
+	img = functions.normalize_input(img = img, normalization = normalization)
+
+	#---------------------------------
+
 	#represent
 	embedding = model.predict(img)[0].tolist()
 

deepface/commons/functions.py

@@ -104,6 +104,57 @@ def detect_face(img, detector_backend = 'opencv', grayscale = False, enforce_det
 			else:
 			  raise ValueError("Face could not be detected. Please confirm that the picture is a face photo or consider to set enforce_detection param to False.")
 
+def normalize_input(img, normalization = 'base'):
+
+	#issue 131 declares that some normalization techniques improves the accuracy
+
+	if normalization == 'base':
+		return img
+	else: #@trevorgribble recommend the following idea
+		
+		img *= 255 #restore input in scale of [0, 255] because it was normalized in scale of  [0, 1] in preprocess_face
+
+		if normalization == 'Facenet':
+			mean, std = img.mean(), img.std()
+			img = (img - mean) / std
+
+		elif normalization == 'v1':
+			#BGR mean subtraction / 255 normalization
+			img[..., 0]-= 131.0912
+			img[..., 1] -= 103.8827
+			img[..., 2] -= 91.4953
+			img = img[..., ::-1]
+			img /= 255
+
+		elif(normalization =="v2"):
+			#RGB mean subtraction / 255 normalization
+			img[..., 0]-= 131.0912
+			img[..., 1] -= 103.8827
+			img[..., 2] -= 91.4953
+			img /= 255
+
+		elif(normalization =="v3"):
+			#BGR mean subtraction normalization
+			img[..., 0]-= 131.0912
+			img[..., 1] -= 103.8827
+			img[..., 2] -= 91.4953
+			img = img[..., ::-1]
+
+		elif(normalization =="v4"):
+			#RGB mean subtraction normalization
+			img[..., 0]-= 131.0912
+			img[..., 1] -= 103.8827
+			img[..., 2] -= 91.4953
+
+		elif(normalization=="v6"):
+			# simply / 127.5 - 1 (similar to facenet 2018 model preprocessing step as @iamrishab posted)
+			img /= 127.5
+			img -= 1
+
+	#-----------------------------
+
+	return img
+
 def preprocess_face(img, target_size=(224, 224), grayscale = False, enforce_detection = True, detector_backend = 'opencv', return_region = False, align = True):
 
 	#img might be path, base64 or numpy array. Convert it to numpy whatever it is.
@@ -155,7 +206,7 @@ def preprocess_face(img, target_size=(224, 224), grayscale = False, enforce_dete
 
 	#normalizing the image pixels
 
-	img_pixels = image.img_to_array(img)
+	img_pixels = image.img_to_array(img) #what this line doing? must?
 	img_pixels = np.expand_dims(img_pixels, axis = 0)
 	img_pixels /= 255 #normalize input in [0, 1]