Merge pull request #497 from shirasael/feature/non-binary-gender-result

Feature/non binary gender result

deepface/DeepFace.py

@@ -294,7 +294,11 @@ def analyze(img_path, actions = ('emotion', 'age', 'gender', 'race') , models =
 		{
 			"region": {'x': 230, 'y': 120, 'w': 36, 'h': 45},
 			"age": 28.66,
-			"gender": "woman",
+			"dominant_gender": "Woman",
+			"gender": {
+				'Woman': 99.99407529830933,
+				'Man': 0.005928758764639497,
+			}
 			"dominant_emotion": "neutral",
 			"emotion": {
 				'sad': 37.65260875225067,
@@ -414,18 +418,25 @@ def analyze(img_path, actions = ('emotion', 'age', 'gender', 'race') , models =
 				resp_obj["age"] = int(apparent_age) #int cast is for the exception - object of type 'float32' is not JSON serializable
 
 			elif action == 'gender':
-				if img_224 is None:
+				try:
-					img_224, region = functions.preprocess_face(img = img_path, target_size = (224, 224), grayscale = False, enforce_detection = enforce_detection, detector_backend = detector_backend, return_region = True)
+					if img_224 is None:
+						img_224, region = functions.preprocess_face(img = img_path, target_size = (224, 224), grayscale = False, enforce_detection = enforce_detection, detector_backend = detector_backend, return_region = True)
 
-				gender_prediction = models['gender'].predict(img_224)[0,:]
+					gender_predictions = models['gender'].predict(img_224)[0,:]
 
-				if np.argmax(gender_prediction) == 0:
+					gender_labels = ["Woman", "Man"]
-					gender = "Woman"
+					resp_obj["gender"] = {}
-				elif np.argmax(gender_prediction) == 1:
-					gender = "Man"
 
-				resp_obj["gender"] = gender
+					for i in range(0, len(gender_labels)):
+						gender_label = gender_labels[i]
+						gender_prediction = 100 * gender_predictions[i]
+						resp_obj["gender"][gender_label] = gender_prediction
 
+					resp_obj["dominant_gender"] = gender_labels[np.argmax(gender_predictions)]
+				except Exception as e:
+					resp_obj["dominant_gender"] = None
+					resp_obj["gender"] = None
+					resp_obj["error"] = e
 			elif action == 'race':
 				if img_224 is None:
 					img_224, region = functions.preprocess_face(img = img_path, target_size = (224, 224), grayscale = False, enforce_detection = enforce_detection, detector_backend = detector_backend, return_region = True) #just emotion model expects grayscale images
@@ -444,7 +455,7 @@ def analyze(img_path, actions = ('emotion', 'age', 'gender', 'race') , models =
 
 			#-----------------------------
 
-			if is_region_set != True:
+			if is_region_set != True and region:
 				resp_obj["region"] = {}
 				is_region_set = True
 				for i, parameter in enumerate(region_labels):
@@ -458,14 +469,8 @@ def analyze(img_path, actions = ('emotion', 'age', 'gender', 'race') , models =
 			return resp_obj
 
 	if bulkProcess == True:
+		return resp_objects
 
-		resp_obj = {}
-
-		for i in range(0, len(resp_objects)):
-			resp_item = resp_objects[i]
-			resp_obj["instance_%d" % (i+1)] = resp_item
-
-		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, normalization = 'base', silent=False):
 

deepface/commons/functions.py

@@ -83,7 +83,7 @@ def load_image(img):
 		img = loadBase64Img(img)
 
 	elif url_img:
-		img = np.array(Image.open(requests.get(img, stream=True).raw))
+		img = np.array(Image.open(requests.get(img, stream=True).raw).convert('RGB'))
 
 	elif exact_image != True: #image path passed as input
 		if os.path.isfile(img) != True:

deepface/extendedmodels/Gender.py

@@ -18,6 +18,7 @@ elif tf_version == 2:
 
 #url = 'https://drive.google.com/uc?id=1wUXRVlbsni2FN9-jkS_f4UTUrm1bRLyk'
 
+
 def loadModel(url = 'https://github.com/serengil/deepface_models/releases/download/v1.0/gender_model_weights.h5'):
 
 	model = VGGFace.baseModel()

tests/test_nonbinary_gender.py

@@ -0,0 +1,35 @@
+from deepface import DeepFace
+
+dataset = [
+    'dataset/img1.jpg',
+	'dataset/img5.jpg',
+	'dataset/img6.jpg',
+    'dataset/img7.jpg',
+    'dataset/img9.jpg',
+    'dataset/img11.jpg',
+    'dataset/img11.jpg',
+]
+
+
+def test_gender_prediction():
+    detectors = ['opencv', 'ssd', 'retinaface', 'mtcnn']  # dlib not tested
+    for detector in detectors:
+        test_gender_prediction_with_detector(detector)
+
+
+def test_gender_prediction_with_detector(detector):
+    results = DeepFace.analyze(dataset, actions=('gender',), detector_backend=detector, prog_bar=False,
+                               enforce_detection=False)
+    for result in results:
+        assert 'gender' in result.keys()
+        assert 'dominant_gender' in result.keys() and result["dominant_gender"] in ["Man", "Woman"]
+        if result["dominant_gender"] == "Man":
+            assert result["gender"]["Man"] > result["gender"]["Woman"]
+        else:
+            assert result["gender"]["Man"] < result["gender"]["Woman"]
+    print(f'detector {detector} passed')
+    return True
+
+
+if __name__ == "__main__":
+    test_gender_prediction()

tests/unit_tests.py

@@ -3,6 +3,7 @@ import os
 import tensorflow as tf
 import cv2
 from deepface import DeepFace
+from tests.test_nonbinary_gender import test_gender_prediction, test_gender_prediction_with_detector
 
 print("-----------------------------------------")
 
@@ -96,7 +97,7 @@ def test_cases():
 	print(demography)
 
 	evaluate(demography["age"] > 20 and demography["age"] < 40)
-	evaluate(demography["gender"] == "Woman")
+	evaluate(demography["dominant_gender"] == "Woman")
 
 	print("-----------------------------------------")
 
@@ -108,13 +109,13 @@ def test_cases():
 
 	#check response is a valid json
 	print("Age: ", demography["age"])
-	print("Gender: ", demography["gender"])
+	print("Gender: ", demography["dominant_gender"])
 	print("Race: ", demography["dominant_race"])
 	print("Emotion: ", demography["dominant_emotion"])
 
 	evaluate(demography.get("age") is not None)
-	evaluate(demography.get("gender") is not None) 
+	evaluate(demography.get("dominant_gender") is not None)
-	evaluate(demography.get("dominant_race") is not None) 
+	evaluate(demography.get("dominant_race") is not None)
 	evaluate(demography.get("dominant_emotion") is not None)
 
 	print("-----------------------------------------")
@@ -123,12 +124,12 @@ def test_cases():
 	demography = DeepFace.analyze(img, ['age', 'gender'])
 
 	print("Age: ", demography.get("age"))
-	print("Gender: ", demography.get("gender"))
+	print("Gender: ", demography.get("dominant_gender"))
 	print("Race: ", demography.get("dominant_race"))
 	print("Emotion: ", demography.get("dominant_emotion"))
 
 	evaluate(demography.get("age") is not None)
-	evaluate(demography.get("gender") is not None)
+	evaluate(demography.get("dominant_gender") is not None)
 	evaluate(demography.get("dominant_race") is None)
 	evaluate(demography.get("dominant_emotion") is None)
 
@@ -151,7 +152,7 @@ def test_cases():
 				distance = round(resp_obj["distance"], 2)
 				threshold = resp_obj["threshold"]
 
-				passed = prediction == result 
+				passed = prediction == result
 
 				evaluate(passed)
 
@@ -206,7 +207,14 @@ def test_cases():
 
 	print("--------------------------")
 
+
+def run_gender_prediction_test():
+	for detector in detectors:
+		evaluate(test_gender_prediction_with_detector(detector))
+
+
 test_cases()
+run_gender_prediction_test()
 
 print("num of test cases run: " + str(num_cases))
 print("succeeded test cases: " + str(succeed_cases))