[英]tensorflow serving produces wrong answer when I export keras model
I tried export my keras model to tensorflow serving and all works well. 我尝试将我的keras模型导出到tensorflow服务,并且一切正常。 What I'm trying to do is to accept a b64 encoded input image string from client and output a True/False value.
我想做的是从客户端接受b64编码的输入图像字符串并输出True / False值。 My keras model outputs 3 values and the first value indicates the degree predicted from model, and I will compare it to another fixed value and export the whole algorithm from taking image string to outputing True/False value to tensorflow serving using RESTful API.
我的keras模型输出3个值,第一个值表示从模型预测的程度,我将其与另一个固定值进行比较,然后将整个算法从获取图像字符串导出到使用RESTful API将True / False值输出到Tensorflow服务。 However, I did not get correct output from my client program.
但是,我没有从客户端程序获得正确的输出。 Long words short, let me show the code
长话短说,让我展示代码
My program to export saved model: 我的程序导出保存的模型:
import tensorflow as tf
from tensorflow.python.saved_model import builder as saved_model_builder
from tensorflow.python.saved_model import tag_constants, signature_constants, signature_def_utils_impl
from keras.models import load_model
from keras.layers import Input
import os
tf.app.flags.DEFINE_string('model_dir', './keras_models',
'''Directory which contains keras models''')
tf.app.flags.DEFINE_string('output_dir', './model_output',
'''Directory where to export the model''')
tf.app.flags.DEFINE_string('model_version', '1',
'''version number of the model''')
tf.app.flags.DEFINE_string('model_file', 'pointer_model.json',
'''json file which contains model architecture''')
tf.app.flags.DEFINE_string('weights_file', 'pointer_model.h5',
'''h5 file that contains model weights''')
FLAGS = tf.app.flags.FLAGS
def preprocess_image(image_buffer):
'''
Preprocess JPEG encoded bytes to 3D floate tensor
:param image_buffer:
:return: 4D image tensor (1, width, height, channels)
'''
image = tf.image.decode_jpeg(image_buffer, channels=3)
image = tf.image.convert_image_dtype(image, dtype=tf.float32)
return image
def main(_):
with tf.Graph().as_default():
serialized_tf_example = tf.placeholder(tf.string, name='input_image')
feature_configs = {
'image/encoded': tf.FixedLenFeature(
shape=[], dtype=tf.string),
}
tf_example = tf.parse_example(serialized_tf_example, feature_configs)
jpegs = tf_example['image/encoded']
images = tf.map_fn(preprocess_image, jpegs, dtype=tf.float32)
images = tf.squeeze(images, [0])
images = tf.expand_dims(images, axis=0)
# now the image shape is [1, ?, ?, 3]
images = tf.image.resize_images(images, tf.constant([224, 224]))
model = load_model('./keras_models/my_model.h5')
x = Input(tensor=images)
y = model(x)
model.summary()
compare_value = tf.Variable(100.0)
bool_out = tf.math.greater(y, compare_value)
bool_out = bool_out[:,0]
bool_out = tf.cast(bool_out, tf.float32)
bool_out = tf.expand_dims(bool_out, axis=0)
final_out = tf.concat([tf.transpose(y), bool_out], axis=0)
init = tf.global_variables_initializer()
with tf.Session() as sess:
sess.run(init)
# predict_tensor_input_info = tf.saved_model.utils.build_tensor_info(jpegs)
# predict_tensor_score_info = tf.saved_model.utils.build_tensor_info(bool_out)
prediction_signature = \
(tf.saved_model.signature_def_utils.predict_signature_def(
inputs={'images': jpegs},
outputs={'scores': final_out}
)
)
export_path = os.path.join(
tf.compat.as_bytes(FLAGS.output_dir),
tf.compat.as_bytes(FLAGS.model_version)
)
builder = saved_model_builder.SavedModelBuilder(export_path)
legacy_init_op = tf.group(tf.tables_initializer(),
name = 'legacy_init_op')
builder.add_meta_graph_and_variables(
sess, [tag_constants.SERVING],
signature_def_map={
signature_constants.DEFAULT_SERVING_SIGNATURE_DEF_KEY:prediction_signature,
},
legacy_init_op = legacy_init_op
)
builder.save()
if __name__ =="__main__":
tf.app.run()
and this is my client program: 这是我的客户程序:
import base64
import requests
import json
import argparse
import time
from glob import glob
image_path = glob('./segmented_image/*.jpg')
for i in range(len(image_path)):
input_image = open(image_path[i], 'rb').read()
encoded_input_string = base64.b64encode(input_image)
input_string = encoded_input_string.decode('utf-8')
# input_image_recover = base64.b64decode(input_string)
# with open('recovered_image.jpg', 'wb') as output_file:
# output_file.write(input_image_recover)
#
# print('Base64 encoded string: ' + input_string[:10] + '...' + input_string[-10:])
instance = [{"b64": input_string}]
data = json.dumps({"instances": instance})
print(data[:30]+ '...' + data[-10:])
json_response = requests.post('http://localhost:8501/v1/models/pointer_model:predict',
data=data)
print(json_response.text)
end_time = time.time()
The output from json_response.text is like: json_response.text的输出类似于:
{"instances": [{"b64": "/9j/4A...Y//9k="}]}
{
"predictions": [[-0.00015692], [-0.000967527], [0.000567942], [0.0]
]
}
{"instances": [{"b64": "/9j/4A...if/9k="}]}
{
"predictions": [[-0.000157582], [-0.000998327], [0.000598866], [0.0]
]
}
...... ......
The first 3 values in prediction key is supposed to be degree, and x,y coordinates in an image which should be hundreds value... the last value is the True/False value casted to float32 comparing with 100.0 预测键中的前三个值应该是度,图像中的x,y坐标应该是几百个值...最后一个值是将float / 32的True / False值与100.0进行比较
Ok.. and last, I have also tested my model using model.predict, which gives correct answer... 好的..最后,我还使用model.predict测试了我的模型,它给出了正确的答案...
Now I'm completely confused. 现在我完全感到困惑。 Can someone tell me where is wrong with my code?
有人可以告诉我我的代码哪里出问题了吗?
use my script to export in tensorflow serving format 使用我的脚本以TensorFlow服务格式导出
import sys
from keras.models import load_model
import tensorflow as tf
from keras import backend as K
from tensorflow.python.framework import graph_util
from tensorflow.python.framework import graph_io
from tensorflow.python.saved_model import signature_constants
from tensorflow.python.saved_model import tag_constants
K.set_learning_phase(0)
K.set_image_data_format('channels_last')
INPUT_MODEL = sys.argv[1]
NUMBER_OF_OUTPUTS = 1
OUTPUT_NODE_PREFIX = 'output_node'
OUTPUT_FOLDER= 'frozen'
OUTPUT_GRAPH = 'frozen_model.pb'
OUTPUT_SERVABLE_FOLDER = sys.argv[2]
INPUT_TENSOR = sys.argv[3]
try:
model = load_model(INPUT_MODEL)
except ValueError as err:
print('Please check the input saved model file')
raise err
output = [None]*NUMBER_OF_OUTPUTS
output_node_names = [None]*NUMBER_OF_OUTPUTS
for i in range(NUMBER_OF_OUTPUTS):
output_node_names[i] = OUTPUT_NODE_PREFIX+str(i)
output[i] = tf.identity(model.outputs[i], name=output_node_names[i])
print('Output Tensor names: ', output_node_names)
sess = K.get_session()
try:
frozen_graph = graph_util.convert_variables_to_constants(sess, sess.graph.as_graph_def(), output_node_names)
graph_io.write_graph(frozen_graph, OUTPUT_FOLDER, OUTPUT_GRAPH, as_text=False)
print(f'Frozen graph ready for inference/serving at {OUTPUT_FOLDER}/{OUTPUT_GRAPH}')
except:
print('Error Occured')
builder = tf.saved_model.builder.SavedModelBuilder(OUTPUT_SERVABLE_FOLDER)
with tf.gfile.GFile(f'{OUTPUT_FOLDER}/{OUTPUT_GRAPH}', "rb") as f:
graph_def = tf.GraphDef()
graph_def.ParseFromString(f.read())
sigs = {}
OUTPUT_TENSOR = output_node_names
with tf.Session(graph=tf.Graph()) as sess:
tf.import_graph_def(graph_def, name="")
g = tf.get_default_graph()
inp = g.get_tensor_by_name(INPUT_TENSOR)
out = g.get_tensor_by_name(OUTPUT_TENSOR[0] + ':0')
sigs[signature_constants.DEFAULT_SERVING_SIGNATURE_DEF_KEY] = \
tf.saved_model.signature_def_utils.predict_signature_def(
{"input": inp}, {"outout": out})
builder.add_meta_graph_and_variables(sess,
[tag_constants.SERVING],
signature_def_map=sigs)
try:
builder.save()
print(f'Model ready for deployment at {OUTPUT_SERVABLE_FOLDER}/saved_model.pb')
print('Prediction signature : ')
print(sigs['serving_default'])
except:
print('Error Occured, please checked frozen graph')
声明:本站的技术帖子网页,遵循CC BY-SA 4.0协议,如果您需要转载,请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.