serving_client调用rpc报错(FasterRCNN)

[wszph]

你好，我在测试提供的目标检测模型提供Faster RCNN Model实例 时，可以正常启动serve端，但是client端报错如下：

WARNING: Logging before InitGoogleLogging() is written to STDERR
E0522 08:31:33.885603   161 config_manager.cpp:217] Not found key in configue: cluster
E0522 08:31:33.885659   161 config_manager.cpp:234] Not found key in configue: split_tag_name
E0522 08:31:33.885700   161 config_manager.cpp:235] Not found key in configue: tag_candidates
E0522 08:31:33.885725   161 config_manager.cpp:202] Not found key in configue: connect_timeout_ms
E0522 08:31:33.885735   161 config_manager.cpp:203] Not found key in configue: rpc_timeout_ms
E0522 08:31:33.885742   161 config_manager.cpp:205] Not found key in configue: hedge_request_timeout_ms
E0522 08:31:33.885751   161 config_manager.cpp:207] Not found key in configue: connect_retry_count
E0522 08:31:33.885767   161 config_manager.cpp:209] Not found key in configue: hedge_fetch_retry_count
E0522 08:31:33.885776   161 config_manager.cpp:211] Not found key in configue: max_connection_per_host
E0522 08:31:33.885790   161 config_manager.cpp:212] Not found key in configue: connection_type
E0522 08:31:33.885798   161 config_manager.cpp:219] Not found key in configue: load_balance_strategy
E0522 08:31:33.885807   161 config_manager.cpp:221] Not found key in configue: cluster_filter_strategy
E0522 08:31:33.885814   161 config_manager.cpp:226] Not found key in configue: protocol
E0522 08:31:33.885821   161 config_manager.cpp:227] Not found key in configue: compress_type
E0522 08:31:33.885829   161 config_manager.cpp:228] Not found key in configue: package_size
E0522 08:31:33.885836   161 config_manager.cpp:230] Not found key in configue: max_channel_per_request
E0522 08:31:33.885844   161 config_manager.cpp:234] Not found key in configue: split_tag_name
E0522 08:31:33.885854   161 config_manager.cpp:235] Not found key in configue: tag_candidates
I0522 08:31:33.891280   161 naming_service_thread.cpp:209] brpc::policy::ListNamingService("127.0.0.1:9494"): added 1

测试代码也是采用实例中的test_client.py文件，麻烦看一下什么原因？

[github-actions]

Message that will be displayed on users' first issue

[barrierye]

这个应该是Client端的正常输出，程序运行的返回值正常吗？

[wszph]

@barrierye 现在Faster RCNN Model实例代码可以正常测试。现在有另外一个问题，我的serving_server_conf.prototxt如下：
feed_var {
name: "image"
alias_name: "image"
is_lod_tensor: false
feed_type: 1
shape: 3
shape: 608
shape: 608
}
feed_var {
name: "im_shape"
alias_name: "im_shape"
is_lod_tensor: false
feed_type: 0
shape: 2
}
fetch_var {
name: "save_infer_model/scale_0.tmp_0"
alias_name: "outputs"
is_lod_tensor: true
fetch_type: 1
shape: -1
}
serving_client_conf.prototxt如下：
feed_var {
name: "image"
alias_name: "image"
is_lod_tensor: false
feed_type: 1
shape: 3
shape: 608
shape: 608
}
feed_var {
name: "im_shape"
alias_name: "im_shape"
is_lod_tensor: false
feed_type: 0
shape: 2
}
fetch_var {
name: "save_infer_model/scale_0.tmp_0"
alias_name: "outputs"
is_lod_tensor: true
fetch_type: 1
shape: -1
}

可以成功启动serving，但写client时，“im_shape”该怎么写？

[bjjwwang]

@barrierye 现在Faster RCNN Model实例代码可以正常测试。现在有另外一个问题，我的serving_server_conf.prototxt如下：
feed_var {
name: "image"
alias_name: "image"
is_lod_tensor: false
feed_type: 1
shape: 3
shape: 608
shape: 608
}
feed_var {
name: "im_shape"
alias_name: "im_shape"
is_lod_tensor: false
feed_type: 0
shape: 2
}
fetch_var {
name: "save_infer_model/scale_0.tmp_0"
alias_name: "outputs"
is_lod_tensor: true
fetch_type: 1
shape: -1
}
serving_client_conf.prototxt如下：
feed_var {
name: "image"
alias_name: "image"
is_lod_tensor: false
feed_type: 1
shape: 3
shape: 608
shape: 608
}
feed_var {
name: "im_shape"
alias_name: "im_shape"
is_lod_tensor: false
feed_type: 0
shape: 2
}
fetch_var {
name: "save_infer_model/scale_0.tmp_0"
alias_name: "outputs"
is_lod_tensor: true
fetch_type: 1
shape: -1
}

可以成功启动serving，但写client时，“im_shape”该怎么写？

我记得imshape没什么影响，预留给图片的长和宽。比如[608, 608]

[wszph]

@barrierye 现在Faster RCNN Model实例代码可以正常测试。现在有另外一个问题，我的serving_server_conf.prototxt如下：
feed_var {
name: "image"
alias_name: "image"
is_lod_tensor: false
feed_type: 1
shape: 3
shape: 608
shape: 608
}
feed_var {
name: "im_shape"
alias_name: "im_shape"
is_lod_tensor: false
feed_type: 0
shape: 2
}
fetch_var {
name: "save_infer_model/scale_0.tmp_0"
alias_name: "outputs"
is_lod_tensor: true
fetch_type: 1
shape: -1
}
serving_client_conf.prototxt如下：
feed_var {
name: "image"
alias_name: "image"
is_lod_tensor: false
feed_type: 1
shape: 3
shape: 608
shape: 608
}
feed_var {
name: "im_shape"
alias_name: "im_shape"
is_lod_tensor: false
feed_type: 0
shape: 2
}
fetch_var {
name: "save_infer_model/scale_0.tmp_0"
alias_name: "outputs"
is_lod_tensor: true
fetch_type: 1
shape: -1
}
可以成功启动serving，但写client时，“im_shape”该怎么写？

我记得imshape没什么影响，预留给图片的长和宽。比如[608, 608]

我在client中尝试用[608, 608]，
serving端会报错：
InvalidArgumentError: Tensor holds the wrong type, it holds int64_t, but desires to be int.
[Hint: Expected valid == true, but received valid:0 != true:1.] at (/paddle/paddle/fluid/framework/tensor_impl.h:33)
[operator < yolo_box > error]

我是用yolov3微调的模型，完整的测试代码如下：
from paddle_serving_client import Client

from paddle_serving_app.reader import Sequential, File2Image, Resize

from paddle_serving_app.reader import RGB2BGR, Transpose, Div, Normalize

import time
import numpy as np

client = Client()

client.load_client_config("yolov3_seq20_client/serving_client_conf.prototxt")

client.connect(["127.0.0.1:9494"])

seq = Sequential([

File2Image(), RGB2BGR(),Resize((608,608)), 

Div(255), Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], False),Transpose((2, 0, 1))

])

start = time.time()

image_file = "HouseFive696.jpg"

#np.array(list(img.shape[1:]) + [1.0])
img = seq(image_file)

fetch_map = client.predict(
feed={
"image": img,
"im_shape": np.array([608,608])
},
fetch=["outputs"])

print("fetch_map",type(fetch_map))

end = time.time()

print(end - start)
麻烦看一下，我该怎么修改client，谢谢

[bjjwwang]

@barrierye 现在Faster RCNN Model实例代码可以正常测试。现在有另外一个问题，我的serving_server_conf.prototxt如下：
feed_var {
name: "image"
alias_name: "image"
is_lod_tensor: false
feed_type: 1
shape: 3
shape: 608
shape: 608
}
feed_var {
name: "im_shape"
alias_name: "im_shape"
is_lod_tensor: false
feed_type: 0
shape: 2
}
fetch_var {
name: "save_infer_model/scale_0.tmp_0"
alias_name: "outputs"
is_lod_tensor: true
fetch_type: 1
shape: -1
}
serving_client_conf.prototxt如下：
feed_var {
name: "image"
alias_name: "image"
is_lod_tensor: false
feed_type: 1
shape: 3
shape: 608
shape: 608
}
feed_var {
name: "im_shape"
alias_name: "im_shape"
is_lod_tensor: false
feed_type: 0
shape: 2
}
fetch_var {
name: "save_infer_model/scale_0.tmp_0"
alias_name: "outputs"
is_lod_tensor: true
fetch_type: 1
shape: -1
}
可以成功启动serving，但写client时，“im_shape”该怎么写？
我记得imshape没什么影响，预留给图片的长和宽。比如[608, 608]

我在client中尝试用[608, 608]，
serving端会报错：
InvalidArgumentError: Tensor holds the wrong type, it holds int64_t, but desires to be int.
[Hint: Expected valid == true, but received valid:0 != true:1.] at (/paddle/paddle/fluid/framework/tensor_impl.h:33)
[operator < yolo_box > error]

我是用yolov3微调的模型，完整的测试代码如下：
from paddle_serving_client import Client

from paddle_serving_app.reader import Sequential, File2Image, Resize

from paddle_serving_app.reader import RGB2BGR, Transpose, Div, Normalize

import time
import numpy as np

client = Client()

client.load_client_config("yolov3_seq20_client/serving_client_conf.prototxt")

client.connect(["127.0.0.1:9494"])

seq = Sequential([

File2Image(), RGB2BGR(),Resize((608,608)), 

Div(255), Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], False),Transpose((2, 0, 1))

])

start = time.time()

image_file = "HouseFive696.jpg"

#np.array(list(img.shape[1:]) + [1.0])
img = seq(image_file)

fetch_map = client.predict(
feed={
"image": img,
"im_shape": np.array([608,608])
},
fetch=["outputs"])

print("fetch_map",type(fetch_map))

end = time.time()

print(end - start)
麻烦看一下，我该怎么修改client，谢谢

可以参考一下 python/examples/faster_rcnn_model/test_client.py