tfserving备忘
2021-03-01 11:26
标签：ict div open response copy 开始 docker default url 启动： docker run -p port:8500 -e MODEL_NAME=ground_segmentation --name tfserving-ground-segmentation -e NVIDIA_VISIBLE_DEVICES=1 -v /model_path_in_machine:/models/ground_segmentation docker_images:tag --per_process_gpu_memory_fraction=0.2 需要注意的是：/model_path_in_machine下应该是版本号，例如0，0下一级目录是saved_model 请求： tfserving备忘 标签：ict div open response copy 开始 docker default url 原文地址：https://www.cnblogs.com/xiaoxiaomajinjiebiji/p/14412698.htmlimport grpc
import numpy as np
import tensorflow as tf
from PIL import Image
from tensorflow_serving.apis import predict_pb2
from tensorflow_serving.apis import prediction_service_pb2_grpc
import cv2
import matplotlib.pyplot as plt
def request_tfserving(inputs, server_url, model_name, signature_name, input_names, output_names):
# 建立连接
channel = grpc.insecure_channel(server_url)
stub = prediction_service_pb2_grpc.PredictionServiceStub(channel)
# 开始设置请求数据
request = predict_pb2.PredictRequest()
request.model_spec.name = model_name # 模型名称
request.model_spec.signature_name = signature_name # 签名名称（默认 serving_default）
# 设置输入数据
for input, input_name in zip(inputs, input_names):
request.inputs[input_name].CopyFrom(tf.make_tensor_proto(input, shape=list(input.shape)))
response = stub.Predict(request, 5.0) # 其中第2个参数为请求的 timeout 时长
res_from_server_np = []
for output_name in output_names:
res_from_server_np.append(tf.make_ndarray(response.outputs[output_name]))
print(np.argmax(res_from_server_np[0][0,:],axis=2).shape)
f, (ax1, ax2) = plt.subplots(1, 2, figsize=(8, 8))
ax1.imshow(input[0,:].astype(np.uint8))
ax2.imshow(np.argmax(res_from_server_np[0][0,:],axis=2))
plt.show()
if __name__ == "__main__":
img_path = ‘/home/ground_train/test/20112201814-1_00000036_img.png‘
img = Image.open(img_path)
img = img.resize((960, 544), Image.NEAREST).convert(‘RGB‘)
img = image.img_to_array(img)
img = np.expand_dims(img, axis=0).astype(‘float32‘)
print(img.shape)
request_tfserving(inputs=[img],#这里一定得是个list
server_url=‘ip:port‘,
model_name=‘ground_segmentation‘,
signature_name=‘serving_default‘,
input_names=[‘input‘],
output_names=[‘output‘])