handling many faces

.gitignore

@@ -17,6 +17,7 @@ deepface/extendedmodels/__pycache__/*
 deepface/subsidiarymodels/__pycache__/*
 deepface/detectors/__pycache__/*
 tests/dataset/*.pkl
+tests/sandbox.ipynb
 .DS_Store
 deepface/.DS_Store
 *.pyc

deepface/commons/functions.py

@@ -27,24 +27,6 @@ elif tf_major_version == 2:
 
 #--------------------------------------------------
 
-def initialize_input(img1_path, img2_path = None):
-
-	if type(img1_path) == list:
-		bulkProcess = True
-		img_list = img1_path.copy()
-	else:
-		bulkProcess = False
-
-		if (
-			(type(img2_path) == str and img2_path != None) #exact image path, base64 image
-			or (isinstance(img2_path, np.ndarray) and img2_path.any()) #numpy array
-		):
-			img_list = [[img1_path, img2_path]]
-		else: #analyze function passes just img1_path
-			img_list = [img1_path]
-
-	return img_list, bulkProcess
-
 def initialize_folder():
 	home = get_deepface_home()
 
@@ -59,6 +41,8 @@ def initialize_folder():
 def get_deepface_home():
 	return str(os.getenv('DEEPFACE_HOME', default=Path.home()))
 
+#--------------------------------------------------
+
 def loadBase64Img(uri):
    encoded_data = uri.split(',')[1]
    nparr = np.fromstring(base64.b64decode(encoded_data), np.uint8)
@@ -93,35 +77,71 @@ def load_image(img):
 
 	return img
 
-def detect_face(img, detector_backend = 'opencv', grayscale = False, enforce_detection = True, align = True):
+#--------------------------------------------------
 
+def extract_faces(img, target_size=(224, 224), detector_backend = 'opencv', grayscale = False, enforce_detection = True, align = True):
+
+	# this is going to store a list of img itself (numpy), it region and confidence
+	extracted_faces = []
+
+	#img might be path, base64 or numpy array. Convert it to numpy whatever it is.
+	img = load_image(img)
 	img_region = [0, 0, img.shape[1], img.shape[0]]
 
-	#----------------------------------------------
-	#people would like to skip detection and alignment if they already have pre-processed images
 	if detector_backend == 'skip':
-		return img, img_region
+		face_objs = [(img, img_region, 0)]
-
-	#----------------------------------------------
-
-	#detector stored in a global variable in FaceDetector object.
-	#this call should be completed very fast because it will return found in memory
-	#it will not build face detector model in each call (consider for loops)
-	face_detector = FaceDetector.build_model(detector_backend)
-
-	try:
-		detected_face, img_region, _ = FaceDetector.detect_face(face_detector, detector_backend, img, align)
-	except: #if detected face shape is (0, 0) and alignment cannot be performed, this block will be run
-		detected_face = None
-
-	if (isinstance(detected_face, np.ndarray)):
-		return detected_face, img_region
 	else:
-		if detected_face == None:
+		face_detector = FaceDetector.build_model(detector_backend)
-			if enforce_detection != True:
+		face_objs = FaceDetector.detect_faces(face_detector, detector_backend, img, align)
-				return img, img_region
+
-			else:
+	# in case of no face found
-				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.")
+	if len(face_objs) == 0 and enforce_detection == True:
+		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.")
+	elif len(face_objs) == 0 and enforce_detection == False:
+		face_objs = [(img, img_region, 0)]
+
+	for current_img, current_region, confidence in face_objs:
+		if current_img.shape[0] > 0 and current_img.shape[1] > 0:
+
+			if grayscale == True:
+				current_img = cv2.cvtColor(current_img, cv2.COLOR_BGR2GRAY)
+
+			# resize and padding
+			if current_img.shape[0] > 0 and current_img.shape[1] > 0:
+				factor_0 = target_size[0] / current_img.shape[0]
+				factor_1 = target_size[1] / current_img.shape[1]
+				factor = min(factor_0, factor_1)
+
+				dsize = (int(current_img.shape[1] * factor), int(current_img.shape[0] * factor))
+				current_img = cv2.resize(current_img, dsize)
+
+				diff_0 = target_size[0] - current_img.shape[0]
+				diff_1 = target_size[1] - current_img.shape[1]
+				if grayscale == False:
+					# Put the base image in the middle of the padded image
+					current_img = np.pad(current_img, ((diff_0 // 2, diff_0 - diff_0 // 2), (diff_1 // 2, diff_1 - diff_1 // 2), (0, 0)), 'constant')
+				else:
+					current_img = np.pad(current_img, ((diff_0 // 2, diff_0 - diff_0 // 2), (diff_1 // 2, diff_1 - diff_1 // 2)), 'constant')
+			
+			#double check: if target image is not still the same size with target.
+			if current_img.shape[0:2] != target_size:
+				current_img = cv2.resize(current_img, target_size)
+			
+			#normalizing the image pixels
+			img_pixels = image.img_to_array(current_img) #what this line doing? must?
+			img_pixels = np.expand_dims(img_pixels, axis = 0)
+			img_pixels /= 255 #normalize input in [0, 1]
+
+			#int cast is for the exception - object of type 'float32' is not JSON serializable
+			region_obj = {"x": int(current_region[0]), "y": int(current_region[1]), "w": int(current_region[2]), "h": int(current_region[3])}
+
+			extracted_face = [img_pixels, region_obj, confidence]
+			extracted_faces.append(extracted_face)
+	
+	if len(extracted_faces) == 0 and enforce_detection == True:
+		raise ValueError("Detected face shape is ", img.shape,". Consider to set enforce_detection argument to False.")
+	
+	return extracted_faces
 
 def normalize_input(img, normalization = 'base'):
 
@@ -169,94 +189,21 @@ def normalize_input(img, normalization = 'base'):
 
 	return img
 
-def preprocess_face(img, target_size=(224, 224), grayscale = False, enforce_detection = True, detector_backend = 'opencv', return_region = False, align = True):
+def find_target_size(model_name):
 
-	#img might be path, base64 or numpy array. Convert it to numpy whatever it is.
+	target_sizes = {
-	img = load_image(img)
+		"VGG-Face": (224, 224),
-	base_img = img.copy()
+		"Facenet": (160, 160),
+		"Facenet512": (160, 160),
+		"OpenFace": (96, 96),
+		"DeepFace": (152, 152),
+		"DeepID": (55, 47), #TODO: might be opposite
+		"Dlib": (150, 150),
+		"ArcFace": (112, 112),
+		"SFace": (112, 112)
+	}
 
-	img, region = detect_face(img = img, detector_backend = detector_backend, grayscale = grayscale, enforce_detection = enforce_detection, align = align)
+	if model_name not in target_sizes.keys():
-
+		raise ValueError(f"unimplemented model name - {model_name}")
-	#--------------------------
+	
-
+	return target_sizes[model_name]
-	if img.shape[0] == 0 or img.shape[1] == 0:
-		if enforce_detection == True:
-			raise ValueError("Detected face shape is ", img.shape,". Consider to set enforce_detection argument to False.")
-		else: #restore base image
-			img = base_img.copy()
-
-	#--------------------------
-
-	#post-processing
-	if grayscale == True:
-		img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
-
-	#---------------------------------------------------
-	#resize image to expected shape
-
-	# img = cv2.resize(img, target_size) #resize causes transformation on base image, adding black pixels to resize will not deform the base image
-
-	if img.shape[0] > 0 and img.shape[1] > 0:
-		factor_0 = target_size[0] / img.shape[0]
-		factor_1 = target_size[1] / img.shape[1]
-		factor = min(factor_0, factor_1)
-
-		dsize = (int(img.shape[1] * factor), int(img.shape[0] * factor))
-		img = cv2.resize(img, dsize)
-
-		# Then pad the other side to the target size by adding black pixels
-		diff_0 = target_size[0] - img.shape[0]
-		diff_1 = target_size[1] - img.shape[1]
-		if grayscale == False:
-			# Put the base image in the middle of the padded image
-			img = np.pad(img, ((diff_0 // 2, diff_0 - diff_0 // 2), (diff_1 // 2, diff_1 - diff_1 // 2), (0, 0)), 'constant')
-		else:
-			img = np.pad(img, ((diff_0 // 2, diff_0 - diff_0 // 2), (diff_1 // 2, diff_1 - diff_1 // 2)), 'constant')
-
-	#------------------------------------------
-
-	#double check: if target image is not still the same size with target.
-	if img.shape[0:2] != target_size:
-		img = cv2.resize(img, target_size)
-
-	#---------------------------------------------------
-
-	#normalizing the image pixels
-
-	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]
-
-	#---------------------------------------------------
-
-	if return_region == True:
-		return img_pixels, region
-	else:
-		return img_pixels
-
-def find_input_shape(model):
-
-	#face recognition models have different size of inputs
-	#my environment returns (None, 224, 224, 3) but some people mentioned that they got [(None, 224, 224, 3)]. I think this is because of version issue.
-
-	input_shape = model.layers[0].input_shape
-
-	if type(input_shape) == list:
-		input_shape = input_shape[0][1:3]
-	else:
-		input_shape = input_shape[1:3]
-
-	#----------------------
-	#issue 289: it seems that tf 2.5 expects you to resize images with (x, y)
-	#whereas its older versions expect (y, x)
-
-	if tf_major_version == 2 and tf_minor_version >= 5:
-		x = input_shape[0]; y = input_shape[1]
-		input_shape = (y, x)
-
-	#----------------------
-
-	if type(input_shape) == list: #issue 197: some people got array here instead of tuple
-		input_shape = tuple(input_shape)
-
-	return input_shape

deepface/commons/realtime.py

@@ -50,7 +50,7 @@ def analysis(db_path, model_name = 'VGG-Face', detector_backend = 'opencv', dist
 
 		#------------------------
 
-		input_shape = functions.find_input_shape(model)
+		input_shape = functions.find_target_size(model_name=model_name)
 		input_shape_x = input_shape[0]; input_shape_y = input_shape[1]
 
 		#tuned thresholds for model and metric pair

tests/unit_tests.py

@@ -1,6 +1,7 @@
 import warnings
 import os
 import tensorflow as tf
+import numpy as np
 import cv2
 from deepface import DeepFace