使用正确图片时人脸识别测试失败问题排查

问题场景 #

刚探索人脸识别技术，运行入门测试代码时出现异常，测试代码如下：

import face_recognition
from PIL import Image
import numpy as np
import base64
from io import BytesIO

def test_face_recognition(image_path):
    # Open the image and ensure it's in RGB format
    img = Image.open(image_path).convert('RGB')

    # Convert image to numpy array
    img_np = np.array(img)

    # Check face detection
    face_locations = face_recognition.face_locations(img_np)
    print(f"Faces detected: {len(face_locations)}")

    if len(face_locations) > 0:
        face_encodings = face_recognition.face_encodings(img_np, known_face_locations=face_locations)
        print(f"Encodings found: {len(face_encodings)}")
    else:
        print("No faces detected.")

# Replace with the path to a local image file
test_face_recognition("test_image.jpg")

测试图片为含有人脸的JPEG文件，代码中已通过img = Image.open(image_path).convert('RGB')将图片转为RGB格式，但仍报错。

错误信息 #

初始错误 #

运行代码后抛出如下错误：

Traceback (most recent call last):
  File "C:\Users\User\source\repos\ballot-box\flask\test_face_recognition.py", line 25, in <module>
    test_face_recognition("test_image.jpg")
  File "C:\Users\User\source\repos\ballot-box\flask\test_face_recognition.py", line 15, in test_face_recognition
    face_locations = face_recognition.face_locations(img_np)
                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\User\source\repos\ballot-box\flask\venv\Lib\site-packages\face_recognition\api.py", line 121, in face_locations
    return [_trim_css_to_bounds(_rect_to_css(face), img.shape) for face in _raw_face_locations(img, number_of_times_to_upsample, model)]
                                                                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\User\source\repos\ballot-box\flask\venv\Lib\site-packages\face_recognition\api.py", line 105, in _raw_face_locations
    return face_detector(img, number_of_times_to_upsample)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: Unsupported image type, must be 8bit gray or RGB image.

修改参数后的错误 #

将传入face_recognition.face_locations()的参数从img_np改为img后，出现新错误：

Traceback (most recent call last):
  File "C:\Users\PauloSantos\source\repos\ballot-box\flask\test_face.py", line 25, in <module>
    test_face_recognition("test_image.jpg")
  File "C:\Users\PauloSantos\source\repos\ballot-box\flask\test_face.py", line 15, in test_face_recognition
    face_locations = face_recognition.face_locations(img)
                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\PauloSantos\source\repos\ballot-box\flask\venv\Lib\site-packages\face_recognition\api.py", line 121, in face_locations
    return [_trim_css_to_bounds(_rect_to_css(face), img.shape) for face in _raw_face_locations(img, number_of_times_to_upsample, model)]
                                                                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\PauloSantos\source\repos\ballot-box\flask\venv\Lib\site-packages\face_recognition\api.py", line 105, in _raw_face_locations
    return face_detector(img, number_of_times_to_upsample)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: __call__(): incompatible function arguments. The following argument types are supported:
    1. (self: _dlib_pybind11.fhog_object_detector, image: numpy.ndarray, upsample_num_times: int = 0) -> _dlib_pybind11.rectangles

Invoked with: <_dlib_pybind11.fhog_object_detector object at 0x000001A8DB8E0370>, <PIL.Image.Image image mode=RGB size=1280x720 at 0x1A8FBD47110>, 1

解决方法 #

问题源于numpy 2.0版本的破坏性更新，导致其与face_recognition库不兼容。执行以下命令将numpy降级至1.26.4即可解决问题：

pip install numpy==1.26.4

内容的提问来源于stack exchange，提问作者Paulo Santos