Hello.

In the function RetinaFace.extract_faces, there is a call to postprocess.alignment_procedure. The arguments left_eye and right_eye have their order switched, which can sometimes lead to faces being aligned upside down.

from retinaface import RetinaFace RuntimeError: module compiled against API version 0xe but this version of numpy is 0xd RuntimeError: module compiled against API version 0xe but this version of numpy is 0xd ImportError: numpy.core.multiarray_umath failed to import ImportError: numpy.core.umath failed to import Traceback (most recent call last): File "", line 1, in File "C:\Users\ROHIT\anaconda3\envs\jarvis\lib\site-packages\retinaface\RetinaFace.py", line 10, in import tensorflow as tf File "C:\Users\ROHIT\anaconda3\envs\jarvis\lib\site-packages\tensorflow_init.py", line 41, in from tensorflow.python.tools import module_util as module_util File "C:\Users\ROHIT\anaconda3\envs\jarvis\lib\site-packages\tensorflow\python_init.py", line 46, in
from tensorflow.python import data File "C:\Users\ROHIT\anaconda3\envs\jarvis\lib\site-packages\tensorflow\python\data_init_.py", line 25, in from tensorflow.python.data import experimental File "C:\Users\ROHIT\anaconda3\envs\jarvis\lib\site-packages\tensorflow\python\data\experimental_init_.py", line 99, in File "C:\Users\ROHIT\anaconda3\envs\jarvis\lib\site-packages\tensorflow\python\data\experimental\service_init_.py", line 140, in from tensorflow.python.data.experimental.ops.data_service_ops import distribute File "C:\Users\ROHIT\anaconda3\envs\jarvis\lib\site-packages\tensorflow\python\data\experimental\ops\data_service_ops.py", line 25, in from tensorflow.python.data.experimental.ops import compression_ops File "C:\Users\ROHIT\anaconda3\envs\jarvis\lib\site-packages\tensorflow\python\data\experimental\ops\compression_ops.py", line 20, in from tensorflow.python.data.util import structure File "C:\Users\ROHIT\anaconda3\envs\jarvis\lib\site-packages\tensorflow\python\data\util\structure.py", line 26, in from tensorflow.python.data.util import nest File "C:\Users\ROHIT\anaconda3\envs\jarvis\lib\site-packages\tensorflow\python\data\util\nest.py", line 40, in from tensorflow.python.framework import sparse_tensor as _sparse_tensor File "C:\Users\ROHIT\anaconda3\envs\jarvis\lib\site-packages\tensorflow\python\framework\sparse_tensor.py", line 28, in from tensorflow.python.framework import constant_op File "C:\Users\ROHIT\anaconda3\envs\jarvis\lib\site-packages\tensorflow\python\framework\constant_op.py", line 29, in from tensorflow.python.eager import execute File "C:\Users\ROHIT\anaconda3\envs\jarvis\lib\site-packages\tensorflow\python\eager\execute.py", line 27, in from tensorflow.python.framework import dtypes File "C:\Users\ROHIT\anaconda3\envs\jarvis\lib\site-packages\tensorflow\python\framework\dtypes.py", line 32, in _np_bfloat16 = _pywrap_bfloat16.TF_bfloat16_type() TypeError: Unable to convert function return value to a Python type! The signature was () -> handle

When calling the postprocess.alignment_procedure function within the RetinaFace.extract_faces function, the order of the arguments is incorrect and changed.

It's a bit annoying that retinaface keeps showing Keras progression bar on every detection. It's especially taxing to Jupyter notebook where it starts to lag after a few hundred faces. I think there should be an option like in deepface where we can set prog_bar=True/False.

!pip install tensorflow==2.2.0（1.9.0）
!pip install h5py==2.10.0
!pip install numpy
!pip install pandas
!pip install gdown
!pip install tqdm
!pip install Pillow
!pip install opencv-python
!pip install opencv-contrib-python
!pip install keras
!pip install Flask
!pip install mtcnn
!pip install lightgbm
!pip install dlib
!pip install retina-face
!pip install deepface

from deepface import DeepFace
from retinaface import RetinaFace

faces = RetinaFace.extract_faces("ai.jpg")

print("There are ", len(faces), " faces in the image")

When I try to execute the above code, I get the following error：

AttributeError                            Traceback (most recent call last)
<ipython-input-4-f1e5ccfe8b11> in <module>
      4 import os
      5 
----> 6 faces = RetinaFace.extract_faces("ai.jpg")
      7 
      8 print("There are ", len(faces), " faces in the image")

/usr/local/lib/python3.6/site-packages/retinaface/RetinaFace.py in extract_faces(img_path, threshold, model, align, allow_upscaling)
    192     #---------------------------
    193 
--> 194     obj = detect_faces(img_path = img, threshold = threshold, model = model, allow_upscaling = allow_upscaling)
    195 
    196     if type(obj) == dict:

/usr/local/lib/python3.6/site-packages/retinaface/RetinaFace.py in detect_faces(img_path, threshold, model, allow_upscaling)
     66 
     67     if model is None:
---> 68         model = build_model()
     69 
     70     #---------------------------

/usr/local/lib/python3.6/site-packages/retinaface/RetinaFace.py in build_model()
     31     if not "model" in globals():
     32 
---> 33         model = tf.function(
     34             retinaface_model.build_model(),
     35             input_signature=(tf.TensorSpec(shape=[None, None, None, 3], dtype=np.float32),)

AttributeError: module 'tensorflow' has no attribute 'function'

Hello, very good recognition results! But very slow. My code: `from retinaface import RetinaFace import cv2 from datetime import datetime

start_time = datetime.now()

img_path = '/home/uba/Desktop/azs/video/test-opencv/frame0-03-56.52.jpg' faces = RetinaFace.detect_faces(img_path) print(faces) print(datetime.now() - start_time) img = cv2.imread(img_path)

for face in faces.keys(): identity = faces[face] facial_area = identity['facial_area'] cv2.rectangle(img, (facial_area[2], facial_area[3]), (facial_area[0], facial_area[1]), (255, 255, 255), 1)

cv2.imshow('', img) cv2.waitKey(0) cv2.destroyAllWindows()`

Computer: AMD Ryzen 5 3600 6-Core Processor GeForce GTX 1660 ram 16+16+8 Ubuntu 20.04

Why so slow ?

In a project working with a colleague, we found that when images were smaller than the default size (1024), e.g., 250 x 250, the runtime was longer than expected because of the upsizing process. However, it appeared to us that the upsizing would not help in detection accuracy. We ended up monkey-patched this proposed change to resize_image, and saw significant improvement to runtime, without sacrificing accuracy. We suggest that the upsizing is probably not necessary (or perhaps making it optional).

I also suggest fixing the order: img_h, img_w, as the first axis of a NumPy image array gives the image height. Even though this does not affect the result in this function, in the future some contributors may get the wrong idea.

Thanks!

Traceback (most recent call last): File "unit-tests.py", line 10, in resp = RetinaFace.detect_faces(img, threshold = 0.5) File "D:\Face_Capture\Server33_Live\RetinaFace\retinaface-master\retinaface\RetinaFace.py", line 54, in detect_faces model = build_model() File "D:\Face_Capture\Server33_Live\RetinaFace\retinaface-master\retinaface\RetinaFace.py", line 30, in build_model retinaface_model.build_model(), File "D:\Face_Capture\Server33_Live\RetinaFace\retinaface-master\retinaface\model\retinaface_model.py", line 486, in build_model ssh_c2_aggr = Conv2D(filters = 256, kernel_size = (3, 3), name = 'ssh_c2_aggr', strides = [1, 1], padding = 'VALID', use_bias = True)(ssh_c2_aggr_pad) File "D:\python_fancp\Anaconda3.7\lib\site-packages\keras\engine\base_layer.py", line 431, in call self.build(unpack_singleton(input_shapes)) File "D:\python_fancp\Anaconda3.7\lib\site-packages\keras\layers\convolutional.py", line 132, in build raise ValueError('The channel dimension of the inputs ' ValueError: The channel dimension of the inputs should be defined. Found None.

Python 3.9.12 (main, Mar 24 2022, 13:02:21) [GCC 11.2.0] on linux Type "help", "copyright", "credits" or "license" for more information.

import retinaface Traceback (most recent call last): File "", line 1, in File "/home/andy/Desktop/img check/retinaface.py", line 1, in from retinaface import RetinaFace ImportError: cannot import name 'RetinaFace' from partially initialized module 'retinaface' (most likely due to a circular import) (/home/andy/Desktop/img check/retinaface.py)

Traceback (most recent call last):
File "face_service.py", line 101, in face_detection    
    bboxes = RetinaFace.detect_faces(image)
File "/usr/local/lib/python3.8/dist-packages/retinaface/RetinaFace.py", line 91, in detect_faces
    net_out = [elt.numpy() for elt in net_out]
File "/usr/local/lib/python3.8/dist-packages/retinaface/RetinaFace.py", line 91, in <listcomp>
    net_out = [elt.numpy() for elt in net_out]
File "/usr/local/lib/python3.8/dist-packages/tensorflow/python/framework/ops.py", line 401, in __getattr__
    self.__getattribute__(name)
AttributeError: 'Tensor' object has no attribute 'numpy'

Traceback (most recent call last): File "", line 1, in File "/home/ubuntu/anaconda3/envs/dev/lib/python3.7/site-packages/retinaface/RetinaFace.py", line 90, in detect_faces net_out = model(im_tensor) File "/home/ubuntu/anaconda3/envs/dev/lib/python3.7/site-packages/tensorflow/python/util/traceback_utils.py", line 153, in error_handler raise e.with_traceback(filtered_tb) from None File "/home/ubuntu/anaconda3/envs/dev/lib/python3.7/site-packages/tensorflow/python/eager/execute.py", line 55, in quick_execute inputs, attrs, num_outputs) tensorflow.python.framework.errors_impl.InternalError: Graph execution error:

Detected at node 'model/bn_data/FusedBatchNormV3' defined at (most recent call last): File "", line 1, in File "/home/ubuntu/anaconda3/envs/dev/lib/python3.7/site-packages/retinaface/RetinaFace.py", line 90, in detect_faces net_out = model(im_tensor) File "/home/ubuntu/anaconda3/envs/dev/lib/python3.7/site-packages/keras/engine/base_layer.py", line 982, in error_handler def call(self, *args, **kwargs): File "/home/ubuntu/anaconda3/envs/dev/lib/python3.7/site-packages/keras/engine/base_layer.py", line 987, in error_handler **kwargs: Keyword arguments to be passed to self.call. File "/home/ubuntu/anaconda3/envs/dev/lib/python3.7/site-packages/keras/engine/base_layer.py", line 1096, in call outputs = call_fn(inputs, *args, **kwargs) File "/home/ubuntu/anaconda3/envs/dev/lib/python3.7/site-packages/keras/utils/traceback_utils.py", line 92, in error_handler return fn(*args, **kwargs) File "/home/ubuntu/anaconda3/envs/dev/lib/python3.7/site-packages/keras/engine/functional.py", line 452, in call inputs, training=training, mask=mask) File "/home/ubuntu/anaconda3/envs/dev/lib/python3.7/site-packages/keras/engine/functional.py", line 589, in _run_internal_graph outputs = node.layer(*args, **kwargs) File "/home/ubuntu/anaconda3/envs/dev/lib/python3.7/site-packages/keras/utils/traceback_utils.py", line 64, in error_handler return fn(*args, **kwargs) File "/home/ubuntu/anaconda3/envs/dev/lib/python3.7/site-packages/keras/engine/base_layer.py", line 1096, in call outputs = call_fn(inputs, *args, **kwargs) File "/home/ubuntu/anaconda3/envs/dev/lib/python3.7/site-packages/keras/utils/traceback_utils.py", line 92, in error_handler return fn(*args, **kwargs) File "/home/ubuntu/anaconda3/envs/dev/lib/python3.7/site-packages/keras/layers/normalization/batch_normalization.py", line 767, in call outputs = self._fused_batch_norm(inputs, training=training) File "/home/ubuntu/anaconda3/envs/dev/lib/python3.7/site-packages/keras/layers/normalization/batch_normalization.py", line 624, in _fused_batch_norm training, train_op, _fused_batch_norm_inference) File "/home/ubuntu/anaconda3/envs/dev/lib/python3.7/site-packages/keras/utils/control_flow_util.py", line 106, in smart_cond pred, true_fn=true_fn, false_fn=false_fn, name=name) File "/home/ubuntu/anaconda3/envs/dev/lib/python3.7/site-packages/keras/layers/normalization/batch_normalization.py", line 613, in _fused_batch_norm_inference data_format=self._data_format) Node: 'model/bn_data/FusedBatchNormV3' cuDNN launch failure : input shape ([1,3,1024,1222]) [[{{node model/bn_data/FusedBatchNormV3}}]] [Op:__inference_function_7605]

• Fixed formatting
• Added "expand_face_area" argument to extract_faces() in Line 206. Now, the function can return the larger area around the face by expand_face_area number of pixels. See issue #37

I wrote a few simple lines to retrieve the extracted faces' original image locations alongside the faces themselves, because when one detects a lot of faces from one image with RetinaFace and uses Deepface to verify all of these faces against some other face reference image, it is very useful to visualize where exactly the verified face was found in the original image. As far as I understand, this functionality in this specific use case was not possible before.

Example semi-pseudo code:

filename = "path_to_image"
faces_and_locations = RetinaFace.extract_faces(filename, return_location = True)

image = Image.open(filename)

model = DeepFace.build_model("Facenet512")
for face, face_location in faces_and_locations:
    verification = DeepFace.verify(img1_path = face, img2_path = "path_to_ref_img", model = model, detector_backend="retinaface", enforce_detection=False)
    
    if verification["verified"] == True:
        draw_rectangle_in_image(image, face_location, "green", width = 5, deepface = False)
    else:
        draw_rectangle_in_image(image, face_location, "red", width = 5, deepface = False)
        
display(image)

So in this example, where I wanted to find David Hasselhoff, we get something like this:

Currently the extracted faces are cropped to a very narrow region around the landmarks, often times cropping out the chin or the hair.

Is is possible to set the bounding box of cropping after alignment/rotation to get larger crops?

Specifically, I would like to have the faces, but with a bit of the surrounding for a timelapse.