人脸 API Python SDK“图像尺寸太小”（PersonGroupPerson add_face_from_stream）

Question

首先，文档here 说“支持JPEG、PNG、GIF（第一帧）和BMP 格式。允许的图像文件大小为1KB 到6MB。”

我正在发送一个大约 1.4 MB 的 .jpg 在我的搜索中，遇到此问题的其他人是自定义形成数据包并遇到了块传输图像的问题。 但是与others 不同的是，我没有形成自己的API 调用，只是将jpg 传递给python sdk。 出了什么问题/我错过了什么？

错误是：

getting image, start time
opening image:  2019_11_30_18_40_21.jpg
time elapsed for capturing image: 8.007975816726685
time elapsed for detecting image: 0.0017137527465820312
appending face found in image
identifying face
time elapsed for identifying image: 0.8008027076721191
Person for face ID e7b2c3fe-6a62-471f-8371-8c1e96608362 is identified in 2019_11_30_18_40_21.jpg with a confidence of 0.68515.
Traceback (most recent call last):
File "./GreeterCam_V0.1 - testing.py", line 116, in <module>
face_client.person_group_person.add_face_from_stream(PERSON_GROUP_ID, face.candidates[0].person_id, image)
File "/home/pi/.local/lib/python3.7/site-packages/azure/cognitiveservices/vision/face/operations/_person_group_person_operations.py", line 785, in add_face_from_stream
raise models.APIErrorException(self._deserialize, response)
azure.cognitiveservices.vision.face.models._models_py3.APIErrorException: (InvalidImageSize) Image size is too small.  

我的源代码是：

if __name__ == '__main__':
    FRAMES_PER_SECOND = 0.13
    ENDPOINT = os.environ['COGNITIVE_SERVICE_ENDPOINT']
    KEY = os.environ['COGNITIVE_SERVICE_KEY']
    face_client = FaceClient(ENDPOINT, CognitiveServicesCredentials(KEY))
    PERSON_GROUP_ID = 'my-unique-person-group'
    #IMAGES_FOLDER = os.path.join(os.path.dirname(os.path.realpath(__file__)))
    #camera = PiCamera()
    #camera.start_preview()
    test_images = [file for file in glob.glob('*.jpg')]
    #webcam = cv2.VideoCapture(0)
    while(True):
        start_time = time.time()
        print('getting image, start time')
        for image_name in test_images:
            image = open(image_name, 'r+b')
            print("opening image: ", image_name)
            time.sleep(5)
            faces = face_client.face.detect_with_stream(image)     
            #image = open(os.path.join(IMAGES_FOLDER, imageName), 'r+b')
            face_ids = []
            time1 = time.time()
            print('time elapsed for capturing image: ' + str(time1-start_time))
            # detect faces in image

            time2 = time.time()
            print('time elapsed for detecting image: ' + str(time2-time1))
            for face in faces:
                print('appending face found in image')
                face_ids.append(face.face_id)
            if face_ids:
                print('identifying face')
                # if there are faces, identify person matching face
                results = face_client.face.identify(face_ids, PERSON_GROUP_ID)
                time3 = time.time()
                print('time elapsed for identifying image: ' + str(time3-time2))
                name = 'person-created-' + str(time.strftime("%Y_%m_%d_%H_%M_%S"))
                if not results:
                    #if there are no matching persons, make a new person and add face
                    print('No person in the person group for faces from {}.'.format(imageName))
                    new_person = face_client.person_group_person.create(PERSON_GROUP_ID, name)
                    face_client.person_group_person.add_face_from_stream(PERSON_GROUP_ID, new_person.person_id, image)
                    time4 = time.time()
                    print('time elapsed for creating new person: ' + str(time4-time3))
                    print('New Person Created: {}'.format(new_person.person_id))
                for face in results:
                    if not face.candidates:
                        new_person = face_client.person_group_person.create(PERSON_GROUP_ID, name)
                        face_client.person_group_person.add_face_from_stream(PERSON_GROUP_ID, new_person.person_id, image)
                    else:
                        #add face to person if match was found
                        print('Person for face ID {} is identified in {} with a confidence of {}.'.format(face.face_id, os.path.basename(image.name), face.candidates[0].confidence)) # Get topmost confidence score
                        face_client.person_group_person.add_face_from_stream(PERSON_GROUP_ID, face.candidates[0].person_id, image)
                        time4 = time.time()
                        print('time elapsed for creating new person: ' + str(time4-time3))   

这也是在 pi 3B(+?) 上的 Raspbian 上

Answer 1

我在我这边运行你的代码并得到同样的错误。似乎代码中的image 参数有问题：

face_client.person_group_person.add_face_from_stream(PERSON_GROUP_ID, face.candidates[0].person_id, image)

在阶段：

#如果找到匹配则添加人脸

当我将此行代码更改为：

face_client.person_group_person.add_face_from_stream(PERSON_GROUP_ID, face.candidates[0].person_id, open(image_name,"r+b"))

问题已解决，人脸已成功添加到人（此人之前有 1 张人脸）：

希望对你有帮助。

Answer 2

我也遇到过这个。这是因为当你在detect_with_stream中使用时，流已经被读取了

您可以转到image.seek(0) 或关闭图像并重新打开它 - 但寻求是更好的解决方案。

Answer 3

我遇到了同样的错误，因为我在识别之前打开了照片。所以我删除了打开并且代码工作了。