Doesn't find any face in webcam images sent as base64 over API

• face_recognition version: 0.1.0
• Python version: 3.6
• Operating System: Windows

Description

Images taken with camera and internet images are easily and quickly recognized even when still uploaded as base64 images, but images taken with the webcam don't come up with any face with the default 'hog' mode until it is changed to the convolutional neural networks (cnn) which is a lot slower, takes about 7 - 10 secs to process one picture

What I Did

This is my django class view:

class WhoInPhoto(APIView):

def post(self, request):
photo = request.data.get('photo', '')
photo = photo.split(',')[1]
all_user_faces = UserImage.objects.all()
known_faces = [face_recognition.face_encodings(face_recognition.load_image_file(img.facepicture.path))[0] for img in all_user_faces]
unknown_face = get_face_encoding_from_base64(photo)
if len(unknown_face) == 0:
return JsonResponse({'status': False, 'message': 'Face not found in photo uploaded'}, safe=False)
results = face_recognition.compare_faces(known_faces, unknown_face, tolerance=0.7)
if True in results:
user = all_user_faces[results.index(True)].user
return JsonResponse({'status': True, 'message': 'Are you {0} ?'.format(user.get_full_name())}, safe=False)
else:
return JsonResponse({'status': False, 'message': 'Face not found in database'}, safe=False)

Then the get_face_encoding function:

def get_face_encoding_from_base64(base64String):
try:
os.mkdir(os.path.join(MEDIA_ROOT, 'tmp'))
except FileExistsError:
pass

IMAGE = Image.open(BytesIO(b64decode(base64String)))
IMAGE.save(os.path.join(os.path.join(MEDIA_ROOT, 'tmp'), 'temp_encoding.jpg'), IMAGE.format)

image = face_recognition.load_image_file(os.path.join(os.path.join(MEDIA_ROOT, 'tmp'), 'temp_encoding.jpg'))

os.unlink(os.path.join(os.path.join(MEDIA_ROOT, 'tmp'), 'temp_encoding.jpg'))

face_locations = face_recognition.face_locations(image, number_of_times_to_upsample=0, model="cnn")
# face_locations = face_recognition.face_locations(image)
print(face_locations)

face_encodings = face_recognition.face_encodings(image, face_locations, 2)

if len(face_encodings) > 0:
return face_encodings[0]
else:
return []

Start with the Django WhoInPhoto view and get_face_encoding_from_base64 function, then reproduce the webcam base64 input while comparing the default HOG face_locations call with the shown CNN call. Check how the decoded and saved image is passed to face_recognition.load_image_file and face_locations; done means the reported webcam images are reliably detected with the intended model and processing path.