Неправильный ответ GRPC от сервера tenorflow - PullRequest
Новая
       30

Неправильный ответ GRPC от сервера tenorflow

0 голосов
/ 10 мая 2019

Я пытаюсь обслуживать API обнаружения объектов tenorflow, но мой клиент grpc и клиенты Rest дают разные результаты для одного и того же изображения.Я не знаю почему.Ответ GRPC:

[2.8745510860517243e-08, 2.476179972177306e-08, 1.955560691158098e-08, 1.1536828381508712e-08, 1.0335512889980691e-08, 9.396929801, 0969393829016.33630348190195e-09, 5.411928682974576e-09, 4.907114270480406e-09, 4.862884761536179e-09, 4.271269560263136e-09, 3.803687587122795e-09, 3.550610694347258e 0930, 093, 098, 098, 098, 098, 098, 09, 28, 09, 09.3, 09.3, 09.3, 09.3.2.471733706599366e-09, 2.4317983182697844e-09, 2.048162306422796e-09]

В то время как ответ клиента REST:

[0.996831, 0.000675639, 0.000323685, 0.000144264, 0.000144642, 0,000134516, 0,000104812, 0,000104108, 9.99449e-05, 8.9907e-05, 8.72486e-05, 6.28879e-05, 6.16111e-05, 6.06435e-05, 5.47078e-05, 4.88681e-05, 4.87645e-05, 4.73167e-05, 4.13763e-05, 4.01956e-05]

Ясно, что клиент GRPC не смог ничего обнаружить.Вот мой GRPC-клиент 

from tensorflow.core.framework import tensor_pb2  
from tensorflow.core.framework import tensor_shape_pb2  
from tensorflow.core.framework import types_pb2

from grpc.beta import implementations
from tensorflow_serving.apis import predict_pb2  
from tensorflow_serving.apis import prediction_service_pb2

import helper

parser = argparse.ArgumentParser(description='incetion grpc client flags.')
parser.add_argument('--host', default='0.0.0.0', help='inception serving host')
parser.add_argument('--port', default='8500', help='inception serving port')
parser.add_argument('--image', dest='image', type=str,
                        help='Path to the jpeg image directory')
FLAGS = parser.parse_args()

def main(file):  
  print("\n\ninput file {}...\n".format(file))
  channel = implementations.insecure_channel(FLAGS.host, int(FLAGS.port))
  stub = prediction_service_pb2.beta_create_PredictionService_stub(channel)

  request = predict_pb2.PredictRequest()
  request.model_spec.name = 'vedNet'
  # request.model_spec.signature_name = 'serving_default'

  img = cv2.imread(file).astype(np.uint8)
  tensor_shape = [1]+list(img.shape)
  dims = [tensor_shape_pb2.TensorShapeProto.Dim(size=dim) for dim in tensor_shape]  
  tensor_shape = tensor_shape_pb2.TensorShapeProto(dim=dims) 

  tensor = tensor_pb2.TensorProto(  
                dtype=types_pb2.DT_UINT8,
                tensor_shape=tensor_shape,
                float_val=list(img.reshape(-1)))
  request.inputs['inputs'].CopyFrom(tensor)  

  resp = stub.Predict(request, 30.0)
  f.close()

  # print(resp.outputs['detection_scores'].float_val)
  print(resp.outputs['detection_scores'].float_val)

if __name__ == '__main__':
  main(FLAGS.image)

и REST-клиент: 

from object_detection.utils import plot_util
from PIL import Image

def pre_process(image_path):
    image = Image.open(image_path).convert("RGB")
    image_np = plot_util.load_image_into_numpy_array(image)

    image_tensor = np.expand_dims(image_np, 0)
    formatted_json_input = json.dumps({"signature_name": "serving_default", "instances": image_tensor.tolist()})

    return formatted_json_input

if __name__ == '__main__':
    parser = argparse.ArgumentParser(description='Performs call to the tensorflow-serving REST API.')
    parser.add_argument('--server_url', dest='server_url', type=str, required=True,
                        help='URL of the tensorflow-serving accepting API call. '
                             'e.g. http://localhost:8501/v1/models/vedNet:predict')
    parser.add_argument('--image_path', dest='image_path', type=str,
                        help='Path to the jpeg image')
    args = parser.parse_args()

    server_url = args.server_url
    image_path = args.image_path

    print("\n\nPre-processing input file {}...\n".format(image_path))
    formatted_json_input = pre_process(image_path)

    headers = {"content-type": "application/json"}
    server_response = requests.post(server_url, data=formatted_json_input, headers=headers)

    response = json.loads(server_response.text)
    output_dict = response['predictions'][0]

    print(output_dict['detection_scores'])

Любая помощь, пожалуйста ...

...