diff --git a/demo/serving/Classification_vgg11_imagenet/README.md b/demo/serving/Classification_vgg11_imagenet/README.md
deleted file mode 100644
index ace0b1a0e58ebe4147f45a04b39606a83e0bb301..0000000000000000000000000000000000000000
--- a/demo/serving/Classification_vgg11_imagenet/README.md
+++ /dev/null
@@ -1,21 +0,0 @@
-## 数据格式  
-input: {files: {"image": [file_1, file_2, ...]}}  
-output: {"results":[result_1, result_2, ...]}  
-
-## Serving快速启动命令  
-```shell
-$ hub serving start -m vgg11_imagenet  
-```  
-
-## python脚本  
-```shell
-$ python vgg11_imagenet_serving_demo.py  
-```  
-
-## 结果示例
-```python
-{  
-    "results": "[[{'Egyptian cat': 0.540287435054779}], [{'daisy': 0.9976677298545837}]]"  
-}  
-```  
-结果含有生成图片的base64编码，可提取生成图片，示例python脚本生成图片位置为当前目录下的output文件夹下。
diff --git a/demo/serving/GAN_stgan_celeba/README.md b/demo/serving/GAN_stgan_celeba/README.md
deleted file mode 100644
index 411e86ad47202cc36644d6f4e8a5a1b80e2405e6..0000000000000000000000000000000000000000
--- a/demo/serving/GAN_stgan_celeba/README.md
+++ /dev/null
@@ -1,47 +0,0 @@
-## 数据格式  
-#### 模型所需参数可通过嵌套字典形式传递
-input: {files: {"image": [file_1, file_2, ...]}, data: {...}}  
-output: {"results":[result_1, result_2, ...]}  
-
-## Serving快速启动命令  
-```shell  
-$ hub serving start -m stgan_celeba  
-```  
-
-## python脚本  
-```shell
-$ python yolov3_coco2017_serving_demo.py
-```
-
-## 结果示例  
-```python
-[  
-    {  
-        "path": "cat.jpg",  
-        "data": [  
-            {  
-                "left": 322.2323,  
-                "right": 1420.4119,  
-                "top": 208.81363,  
-                "bottom": 996.04395,  
-                "label": "cat",  
-                "confidence": 0.9289875  
-            }  
-        ]  
-    },  
-    {  
-        "path": "dog.jpg",  
-        "data": [  
-            {  
-                "left": 204.74722,  
-                "right": 746.02637,  
-                "top": 122.793274,  
-                "bottom": 566.6292,  
-                "label": "dog",  
-                "confidence": 0.86698055  
-            }  
-        ]  
-    }  
-]  
-```
-结果含有生成图片的base64编码，可提取生成图片，示例python脚本生成图片位置为当前目录下的output文件夹下。
diff --git a/demo/serving/Language_Model_lm_lstm/README.md b/demo/serving/Language_Model_lm_lstm/README.md
deleted file mode 100644
index c8a4fc1acada6bfeead62211c7dafb085e7284b3..0000000000000000000000000000000000000000
--- a/demo/serving/Language_Model_lm_lstm/README.md
+++ /dev/null
@@ -1,29 +0,0 @@
-## 数据格式  
-input: {"text": [text_1, text_2, ...]}  
-output: {"results":[result_1, result_2, ...]}  
-
-## Serving快速启动命令  
-```shell
-$ hub serving start -m lm_lstm  
-```
-
-## python脚本  
-```shell
-$ python lm_lstm_serving_demo.py
-```
-
-## 结果示例
-```python
-{  
-    "results": [  
-        {  
-            "perplexity": 4.584166451916099,  
-            "text": "the plant which is owned by <unk> & <unk> co. was under            contract with <unk> to make the cigarette filter"  
-        },  
-        {  
-            "perplexity": 6.038358983397484,  
-            "text": "more common <unk> fibers are <unk> and are more easily            rejected by the body dr. <unk> explained"  
-        }  
-    ]  
-}  
-```  
diff --git a/demo/serving/Language_Model_lm_lstm/lm_lstm_serving_demo.py b/demo/serving/Language_Model_lm_lstm/lm_lstm_serving_demo.py
deleted file mode 100644
index 9142b826e36ce5d8364e82c315a15f446bc2623f..0000000000000000000000000000000000000000
--- a/demo/serving/Language_Model_lm_lstm/lm_lstm_serving_demo.py
+++ /dev/null
@@ -1,17 +0,0 @@
-# coding: utf8
-import requests
-import json
-
-if __name__ == "__main__":
-    # 指定用于用于预测的文本并生成字典{"text": [text_1, text_2, ... ]}
-    text_list = [
-        "the plant which is owned by <unk> & <unk> co. was under contract with <unk> to make the cigarette filter",
-        "more common <unk> fibers are <unk> and are more easily rejected by the body dr. <unk> explained"
-    ]
-    text = {"text": text_list}
-    # 指定预测方法为lm_lstm并发送post请求
-    url = "http://127.0.0.1:8866/predict/text/lm_lstm"
-    r = requests.post(url=url, data=text)
-
-    # 打印预测结果
-    print(json.dumps(r.json(), indent=4, ensure_ascii=False))
diff --git a/demo/serving/Lexical_Analysis_lac/README.md b/demo/serving/Lexical_Analysis_lac/README.md
deleted file mode 100644
index e1eaf4553ece353c9ffebd952c6ed47f4864c5f9..0000000000000000000000000000000000000000
--- a/demo/serving/Lexical_Analysis_lac/README.md
+++ /dev/null
@@ -1,57 +0,0 @@
-## 数据格式  
-input: {"text": [text_1, text_2, ...]}  
-output: {"results":[result_1, result_2, ...]}  
-
-## Serving快速启动命令  
-```shell  
-$ hub serving start -m lac  
-```  
-
-## python脚本  
-#### 不携带用户自定义词典  
-```shell  
-$ python lac_no_dict_serving_demo.py  
-```  
-
-#### 携带用户自定义词典  
-```shell  
-$ python lac_with_dict_serving_demo.py  
-```  
-
-## 结果示例  
-```python
-{  
-    "results": [  
-        {  
-            "tag": [  
-                "TIME",  
-                "v",  
-                "q",  
-                "n"  
-            ],  
-            "word": [  
-                "今天",  
-                "是",  
-                "个",  
-                "好日子"  
-            ]  
-        },  
-        {  
-            "tag": [  
-                "n",  
-                "v",  
-                "TIME",  
-                "v",  
-                "v"  
-            ],  
-            "word": [  
-                "天气预报",  
-                "说",  
-                "今天",  
-                "要",  
-                "下雨"  
-            ]  
-        }  
-    ]  
-}  
-```
diff --git a/demo/serving/Object_Detection_yolov3_coco2017/README.md b/demo/serving/Object_Detection_yolov3_coco2017/README.md
deleted file mode 100644
index 5bf78622719204b88a44e9b8064b88110350db38..0000000000000000000000000000000000000000
--- a/demo/serving/Object_Detection_yolov3_coco2017/README.md
+++ /dev/null
@@ -1,46 +0,0 @@
-## 数据格式  
-input: {files: {"image": [file_1, file_2, ...]}}  
-output: {"results":[result_1, result_2, ...]}  
-
-## Serving快速启动命令  
-```shell  
-$ hub serving start -m yolov3_coco2017  
-```
-
-## python脚本  
-```shell
-$ python yolov3_coco2017_serving_demo.py  
-```
-
-## 结果示例  
-```python
-[  
-    {  
-        "path": "cat.jpg",  
-        "data": [  
-            {  
-                "left": 322.2323,  
-                "right": 1420.4119,  
-                "top": 208.81363,  
-                "bottom": 996.04395,  
-                "label": "cat",  
-                "confidence": 0.9289875  
-            }  
-        ]  
-    },  
-    {  
-        "path": "dog.jpg",  
-        "data": [  
-            {  
-                "left": 204.74722,  
-                "right": 746.02637,  
-                "top": 122.793274,  
-                "bottom": 566.6292,  
-                "label": "dog",  
-                "confidence": 0.86698055  
-            }  
-        ]  
-    }  
-]  
-```
-结果含有生成图片的base64编码，可提取生成图片，示例python脚本生成图片位置为当前目录下的output文件夹下。
diff --git a/demo/serving/README.md b/demo/serving/README.md
index 079b348917a557436c41179c253921c54bac8cfa..11467beb57042d08aa28fb6994fce21fad45de64 100644
--- a/demo/serving/README.md
+++ b/demo/serving/README.md
@@ -1,43 +1,54 @@
 # PaddleHub-Serving
 ## 1. 简介
-利用PaddleHub-Serving可以完成模型服务化部署，主要包括利用Bert as Service实现embedding服务化，利用预测模型实现预测服务化。
+### 1.1 什么是PaddleHub-Serving
+PaddleHub-Serving是基于PaddleHub的一键模型服务部署工具，能够通过简单的Hub命令行工具轻松启动一个模型预测在线服务。
 
-## 2. Bert as Service
-* [Bert as Service介绍与示例](bert_service)
+### 1.2 为什么使用PaddleHub-Serving
+使用PaddleHub能够完成预训练模型的管理和一键预测，但开发者还经常面临将模型部署上线以对外提供服务的需求，而利用PaddleHub-Serving可便捷的将模型部署上线，开发者只需要关注如何处理输入数据和输出结果即可。
 
-该示例展示了利用Bert as Service进行远程embedding服务化部署和在线预测，获取文本embedding结果。
+### 1.3 PaddleHub-Serving的组成
+PaddleHub-Serving主要包括利用Bert Service实现embedding服务化，以及利用预测模型实现预训练模型预测服务化，未来还将支持开发者利用PaddleHub自己finetune后的模型服务化。
 
-## 3. Serving
-模型预测服务化有以下示例：  
+## 2. Bert Service
+Bert Service是基于Paddle Serving框架的快速部署模型远程计算服务方案，可将embedding过程通过调用API接口的方式实现，减少了对机器资源的依赖。使用PaddleHub可在服务器上一键部署`Bert Service`服务，在另外的普通机器上通过客户端接口即可轻松的获取文本对应的embedding数据。
 
-* [图像分类-基于vgg11_imagent](./Classification_vgg11_imagenet)  
+关于Bert Service的具体信息和示例请参阅[Bert Service](./bert_service)
+
+该示例展示了利用Bert Service进行远程embedding服务化部署和在线预测，并获取文本embedding结果。
+
+## 3. PaddleHub-Serving
+预训练模型预测服务化能够一键完成模型服务部署，包括以下示例：  
+
+* [图像分类-基于vgg11_imagent](serving/classification_vgg11_imagenet)  
 
 该示例展示了利用vgg11_imagent完成图像分类服务化部署和在线预测，获取图像分类结果。
 
-* [图像生成-基于stgan_celeba](./GAN_stgan_celeba)  
+* [图像生成-基于stgan_celeba](serving/GAN_stgan_celeba)  
 
 该示例展示了利用stgan_celeba生成图像服务化部署和在线预测，获取指定风格的生成图像。
 
-* [英文词法分析-基于lm_lstm](./Language_Model_lm_lstm)
+* [文本审核-基于porn_detection_lstm](serving/text_censorship_porn_detection_lstm)  
 
-该示例展示了利用lm_lstm完成英文语法分析服务化部署和在线预测，获取文本的流利程度。
+该示例展示了利用porn_detection_lstm完成中文文本黄色敏感信息鉴定的服务化部署和在线预测，获取文本是否敏感及其置信度。
 
-* [中文词法分析-基于lac](./Lexical_Analysis_lac)
+* [中文词法分析-基于lac](serving/lexical_analysis_lac)
 
 该示例展示了利用lac完成中文文本分词服务化部署和在线预测，获取文本的分词结果，并可通过用户自定义词典干预分词结果。
 
-* [目标检测-基于yolov3_coco2017](./Object_Detection_yolov3_coco2017)  
+* [目标检测-基于yolov3_darknet53_coco2017](serving/object_detection_yolov3_darknet53_coco2017)  
 
-该示例展示了利用yolov3_coco2017完成目标检测服务化部署和在线预测，获取检测结果和覆盖识别框的图片。
+该示例展示了利用yolov3_darknet53_coco2017完成目标检测服务化部署和在线预测，获取检测结果和覆盖识别框的图片。
 
-* [中文语义分析-基于simnet_bow](./Semantic_Model_simnet_bow)
+* [中文语义分析-基于simnet_bow](serving/semantic_model_simnet_bow)
 
 该示例展示了利用simnet_bow完成中文文本相似度检测服务化部署和在线预测，获取文本的相似程度。  
 
-* [图像分割-基于deeplabv3p_xception65_humanseg](./Semantic_Segmentation_deeplabv3p_xception65_humanseg)
+* [图像分割-基于deeplabv3p_xception65_humanseg](serving/semantic_segmentation_deeplabv3p_xception65_humanseg)
 
 该示例展示了利用deeplabv3p_xception65_humanseg完成图像分割服务化部署和在线预测，获取识别结果和分割后的图像。
 
-* [中文情感分析-基于senta_lstm](./Sentiment_Analysis_senta_lstm)
+* [中文情感分析-基于simnet_bow](serving/semantic_model_simnet_bow)
 
 该示例展示了利用senta_lstm完成中文文本情感分析服务化部署和在线预测，获取文本的情感分析结果。
+
+关于Paddle-Serving一键模型部署功能的具体信息请参与[serving](./serving)
diff --git a/demo/serving/Semantic_Model_simnet_bow/README.md b/demo/serving/Semantic_Model_simnet_bow/README.md
deleted file mode 100644
index 7a85028bbb3a00ef726ccd996476eb5c2552f83c..0000000000000000000000000000000000000000
--- a/demo/serving/Semantic_Model_simnet_bow/README.md
+++ /dev/null
@@ -1,36 +0,0 @@
-## 数据格式  
-input: {"text1": [text_a1, text_a2, ...], "text2": [text_b1, text_b2, ...]}  
-output: {"results":[result_1, result_2, ...]}  
-
-## Serving快速启动命令  
-```shell
-$ hub serving start -m simnet_bow  
-```
-
-## python脚本  
-```shell
-$ python simnet_bow_serving_demo.py  
-```
-
-## 结果示例  
-```python
-{  
-    "results": [  
-        {  
-            "similarity": 0.8445,  
-            "text_1": "这道题太难了",  
-            "text_2": "这道题是上一年的考题"  
-        },  
-        {  
-            "similarity": 0.9275,  
-            "text_1": "这道题太难了",  
-            "text_2": "这道题不简单"  
-        },  
-        {  
-            "similarity": 0.9083,  
-            "text_1": "这道题太难了",  
-            "text_2": "这道题很有意思"  
-        }  
-    ]  
-}  
-```
diff --git a/demo/serving/Semantic_Segmentation_deeplabv3p_xception65_humanseg/README.md b/demo/serving/Semantic_Segmentation_deeplabv3p_xception65_humanseg/README.md
deleted file mode 100644
index 5f5ad302366e09acd52847606f576e5cba3dcad0..0000000000000000000000000000000000000000
--- a/demo/serving/Semantic_Segmentation_deeplabv3p_xception65_humanseg/README.md
+++ /dev/null
@@ -1,24 +0,0 @@
-## 数据格式  
-input: {files: {"image": [file_1, file_2, ...]}}  
-output: {"results":[result_1, result_2, ...]}  
-
-## Serving快速启动命令  
-```shell
-$ hub serving start -m deeplabv3p_xception65_humanseg  
-```  
-
-## python脚本  
-```shell
-$ python deeplabv3p_xception65_humanseg_serving_demo.py
-```
-
-## 结果示例  
-```python
-[  
-    {  
-        "origin": "girl.jpg",  
-        "processed": "humanseg_output/girl_2.png"  
-    }  
-]  
-```
-结果含有生成图片的base64编码，可提取生成图片，示例python脚本生成图片位置为当前目录下的output文件夹中。
diff --git a/demo/serving/Sentiment_Analysis_senta_lstm/README.md b/demo/serving/Sentiment_Analysis_senta_lstm/README.md
deleted file mode 100644
index 6eadf2f2c3827bc4dfd042b8977f7590a898c275..0000000000000000000000000000000000000000
--- a/demo/serving/Sentiment_Analysis_senta_lstm/README.md
+++ /dev/null
@@ -1,35 +0,0 @@
-## 数据格式  
-input: {"text": [text_1, text_2, ...]}  
-output: {"results":[result_1, result_2, ...]}  
-
-## Serving快速启动命令  
-```shell
-$ hub serving start -m senta_lstm  
-```
-
-## python脚本
-``` shell
-$ python senta_lstm_serving_demo.py  
-```  
-
-## 结果示例  
-```python
-{  
-    "results": [  
-        {  
-            "negative_probs": 0.7079,  
-            "positive_probs": 0.2921,  
-            "sentiment_key": "negative",  
-            "sentiment_label": 0,  
-            "text": "我不爱吃甜食"  
-        },  
-        {  
-            "negative_probs": 0.0149,  
-            "positive_probs": 0.9851,  
-            "sentiment_key": "positive",  
-            "sentiment_label": 1,  
-            "text": "我喜欢躺在床上看电影"  
-        }  
-    ]  
-}  
-```
diff --git a/demo/serving/bert_service/README.md b/demo/serving/bert_service/README.md
index 57c769e1bd159245dc55c023dffd968ec4f70335..192950128fedc63752424e2a711f641cb5a89bd4 100644
--- a/demo/serving/bert_service/README.md
+++ b/demo/serving/bert_service/README.md
@@ -18,7 +18,7 @@
 
 * 专注下游深度学习任务，可利用PaddleHub的`Bert Service`大幅减少embedding代码。  
 
-`Bert as Service`具有几个突出的优点：  
+`Bert Service`具有几个突出的优点：  
 
 * 代码精短，易于使用。简单的pip安装方式，服务端仅需一行命令即可启动，客户端仅需一行代码即可获取embedding结果。  
 
@@ -135,29 +135,29 @@ client端接收文本数据，并获取server端返回的模型计算的embeddin
 
 client端利用PaddleHub的语义理解任务将原始文本按照不同模型的数据预处理方案将文本ID化，并生成对应的sentence type、position、input masks数据，将这些信息封装成json数据，通过http协议按照指定的IP端口信息发送至server端，等待并获取模型生成结果。
 ### 5.2 启动
-连接服务端方法原型为：
+服务端类BSClient初始化方法原型为：
 ```python
-def connect(input_text,
-            model_name,
-            max_seq_len=128,
-            show_ids=False,
-            do_lower_case=True,
-            server="127.0.0.1:8866",
-            retry=3)
+BSClient.__init__(self,
+             module_name,
+             server,
+             max_seq_len=20,
+             show_ids=False,
+             do_lower_case=True,
+             retry=3)
+# 获取embedding方法原型为
+BSClient.get_result(self, input_text)  
 ```  
-
 其中各参数说明如下表：  
 
 |参数|说明|类型|样例|  
 |:--:|:--:|:--:|:--:|  
-|input_text|输入文本，要获取embedding的原始文本|二维list类型，内部元素为string类型的文本|[['样例1'],['样例2']]|  
-|model_name|指定使用的模型名称|string|"ernie"|  
+|module_name|指定使用的模型名称|string|"ernie"|  
+|server|要访问的server地址，包括ip地址及端口号|string|"127.0.0.1:8866"|  
 |max_seq_len|计算时的样例长度，样例长度不足时采用补零策略，超出此参数则超出部分会被截断|int|128|  
 |show_ids|是否展现数据预处理后的样例信息，指定为True则显示样例信息，反之则不显示|bool|False|  
 |do_lower_case|是否将英文字母转换成小写，指定为True则将所有英文字母转换为小写，反之则保持原状|bool|True|
-|server|要访问的server地址，包括ip地址及端口号|string|"127.0.0.1:8866"|  
-|retry|连接失败后的最大重试次数|int|3|
-
+|retry|连接失败后的最大重试次数|int|3|  
+|input_text|输入文本，要获取embedding的原始文本|二维list类型，内部元素为string类型的文本|[['样例1'],['样例2']]|
 
 ## 6. Demo
 在这里，我们将展示一个实际场景中可能使用的demo，我们利用PaddleHub在一台GPU机器上部署`ernie_tiny`模型服务，并在另一台CPU机器上尝试访问，获取一首七言绝句的embedding。
@@ -233,7 +233,7 @@ Paddle Inference Server exit successfully!
  browser."，这个页面有什么作用。  
 > A : 这是`BRPC`的内置服务，主要用于查看请求数、资源占用等信息，可对server端性能有大致了解，具体信息可查看[BRPC内置服务](https://github.com/apache/incubator-brpc/blob/master/docs/cn/builtin_service.md)。
 
-> Q : 为什么输入文本的格式为[["文本1"], ["文本2"], ]，而不是["文本1", "文本2", ]？
+> Q : 为什么输入文本的格式为[["文本1"], ["文本2"], ]，而不是["文本1", "文本2", ]？  
 > A : 因为Bert模型可以对一轮对话生成向量表示，例如[["问题1","回答1"],["问题2","回答2"]]，为了防止使用时混乱，每个样本使用一个list表示，一个样本list内部可以是1条string或2条string，如下面的文本：  
 > ```python
 > input_text = [
diff --git a/demo/serving/serving/GAN_stgan_celeba/README.md b/demo/serving/serving/GAN_stgan_celeba/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..45fe9a87523bbeb02a7affc29c70c87579022e87
--- /dev/null
+++ b/demo/serving/serving/GAN_stgan_celeba/README.md
@@ -0,0 +1,67 @@
+# 部署图像生成服务-以stgan_celeba为例
+## 1 简介
+图像生成是指根据预先设置的标签，生成对应图像的过程。stgan_celeba通过在GAN中加入encoder-decoder，可实现人脸属性的转换。关于stgan_celeba的具体信息请参阅[stgan_celeba](https://paddlepaddle.org.cn/hubdetail?name=stgan_celeba&en_category=GANs)。
+
+使用PaddleHub-Serving可以轻松部署一个在线图像生成服务API，可将此API接入自己的web网站，也可接入应用程序，如美图类应用，实现传照片修饰脸的功能。
+
+下面就带领大家使用PaddleHub-Serving，通过简单几步部署一个图像生成服务。
+
+## 2 启动PaddleHub-Serving
+启动命令如下
+```shell
+$ hub serving start -m stgan_celeba
+```
+启动时会显示加载模型过程，启动成功后显示
+```shell
+Loading stgan_celeba successful.
+```
+这样就完成了一个图像生成服务化API的部署，默认端口号为8866。
+
+## 3 测试图像生成在线API
+我们用来测试的样例图片为  
+
+<p align="center">  
+<img src="../img/man.png" width="30%" />  
+</p>  
+
+根据stgan_celeba所需信息，准备的数据包括图像文件和生成图像风格，格式为
+```python
+files = [("image", file_a), ("image", file_b)]
+data = {"info": ["info_a_1, info_a_2", "info_b_1, info_b_2"], "style": ["style_a", "style_b"]}
+```
+注意文件列表每个元素第一个参数为"image"。
+
+info为图像描述，根据示例图像信息，info应为"Male,Black_Hair,Eyeglasses,No_Beard"，即"男性，黑发，戴眼镜，没有胡子"。
+
+image为要生成的图像风格，我们选取"Bald"(秃顶的)作为生成图像的风格。
+
+代码如下
+```python
+>>> # 指定要使用的图片文件并生成列表[("image", img_1), ("image", img_2), ... ]
+>>> file_list = ["../img/man.png"]
+>>> files = [("image", (open(item, "rb"))) for item in file_list]
+>>> # 为每张图片对应指定info和style
+>>> data = {"info": ["Male,Black_Hair,Eyeglasses,No_Beard"], "style": ["Bald"]}
+```
+然后就可以发送请求到图像生成服务API，并得到结果，代码如下
+
+```python
+>>> url = "http://127.0.0.1:8866/predict/image/stgan_celeba"
+>>> r = requests.post(url=url, data=data, files=files)
+```
+stgan_celeba返回的结果包括生成图像的base64编码格式，经过转换可以得到生成图像，代码如下
+```python
+>>> for item in results:
+...     with open(output_path, "wb") as fp:
+...         fp.write(base64.b64decode(item["base64"].split(',')[-1]))
+```
+查看指定输出文件夹，就能看到生成图像了，如图
+
+<p align="center">  
+<img src="./stgan_output/Bald_man.png" width="30%" />  
+</p>  
+
+
+这样我们就完成了对图像生成服务化的部署和测试。
+
+完整的测试代码见[stgan_celeba_serving_demo.py](stgan_celeba_serving_demo.py)。
diff --git a/demo/serving/GAN_stgan_celeba/stgan_celeba_serving_demo.py b/demo/serving/serving/GAN_stgan_celeba/stgan_celeba_serving_demo.py
similarity index 75%
rename from demo/serving/GAN_stgan_celeba/stgan_celeba_serving_demo.py
rename to demo/serving/serving/GAN_stgan_celeba/stgan_celeba_serving_demo.py
index c6b5129c32d639889319a9a4a6c3fc9721fd5dcf..f345ed272f781c5f11c81ea65cc95e18a10f9859 100644
--- a/demo/serving/GAN_stgan_celeba/stgan_celeba_serving_demo.py
+++ b/demo/serving/serving/GAN_stgan_celeba/stgan_celeba_serving_demo.py
@@ -6,20 +6,21 @@ import os
 
 if __name__ == "__main__":
     # 指定要使用的图片文件并生成列表[("image", img_1), ("image", img_2), ... ]
-    file_list = ["../img/woman.png"]
+    file_list = ["../img/man.png"]
     files = [("image", (open(item, "rb"))) for item in file_list]
     # 为每张图片对应指定info和style
-    data = {"info": ["Female,Brown_Hair"], "style": ["Aged"]}
+    data = {"info": ["Male,Black_Hair"], "style": ["Bald"]}
     # 指定图片生成方法为stgan_celeba并发送post请求
     url = "http://127.0.0.1:8866/predict/image/stgan_celeba"
     r = requests.post(url=url, data=data, files=files)
+    print(r.text)
 
     results = eval(r.json()["results"])
     # 保存生成的图片到output文件夹，打印模型输出结果
-    if not os.path.exists("output"):
-        os.mkdir("output")
+    if not os.path.exists("stgan_output"):
+        os.mkdir("stgan_output")
     for item in results:
-        output_path = os.path.join("output", item["path"].split("/")[-1])
+        output_path = os.path.join("stgan_output", item["path"].split("/")[-1])
         with open(output_path, "wb") as fp:
             fp.write(base64.b64decode(item["base64"].split(',')[-1]))
             item.pop("base64")
diff --git a/demo/serving/serving/GAN_stgan_celeba/stgan_output/Bald_man.png b/demo/serving/serving/GAN_stgan_celeba/stgan_output/Bald_man.png
new file mode 100644
index 0000000000000000000000000000000000000000..e5b17a741668a5e1a3d1df7a4fb519b447a36e31
Binary files /dev/null and b/demo/serving/serving/GAN_stgan_celeba/stgan_output/Bald_man.png differ
diff --git a/demo/serving/serving/README.md b/demo/serving/serving/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..03df13d55ce4b3e7c5066dfa9ddbe24fa83b3f45
--- /dev/null
+++ b/demo/serving/serving/README.md
@@ -0,0 +1,230 @@
+# Hub-Serving一键服务部署
+## 1 简介
+### 1.1 什么是PaddleHub-Serving一键服务部署
+PaddleHub-Serving是基于PaddleHub的一键模型服务部署工具，能够通过简单的Hub命令行工具轻松启动一个模型预测在线服务，开发者可利用PaddleHub-Serving快速得到一个预测服务API。
+
+### 1.2 支持模型
+目前PaddleHub-Serving支持PaddleHub所有可直接用于预测的模型进行服务部署，包括`lac`、`senta_bilstm`等nlp类模型，以及`yolov3_coco2017`、`vgg16_imagenet`等cv类模型，未来还将支持开发者使用自己finetune后的模型用于快捷服务部署。
+
+### 1.3 所需环境
+下表是使用PaddleHub-Serving的环境要求及注意事项。  
+
+|项目|建议版本|说明|  
+|:-:|:-:|:-:|  
+|操作系统|Linux/Darwin/Windows|建议使用Linux或Darwin，对多线程启动方式支持性较好|  
+|PaddleHub|>=1.4.0|无|  
+|PaddlePaddle|>=1.6.1|若使用GPU计算，则对应使用PaddlePaddle-gpu版本|  
+
+## 2 使用
+### 2.1 启动Hub-Serving服务端部署
+Hub-Serving有两种启动方式，分别是使用命令行命令启动，以及使用配置文件启动。
+
+#### 2.1.1 命令行命令启动
+启动命令
+```shell
+$ hub serving start --modules [Module1==Version1, Module2==Version2, ...] \
+                    --port XXXX \
+                    --use_gpu \
+                    --use_multiprocess
+```
+
+**参数**：
+
+|参数|用途|  
+|-|-|  
+|--modules/-m|Hub-Serving预安装模型，以多个Module==Version键值对的形式列出<br/>*`当不指定Version时，默认选择最新版本`*|  
+|--port/-p|服务端口，默认为8866|  
+|--use_gpu|使用GPU进行预测，必须安装paddlepaddle-gpu|  
+|--use_multiprocess|是否启用并发方式，默认为单进程方式|  
+
+#### 2.1.2 配置文件启动
+启动命令
+```shell
+$ hub serving start --config config.json
+```
+`config.json`格式如下：  
+
+```json
+{  
+  "modules_info": [  
+    {  
+       "module": "MODULE_NAME_1",
+       "version": "MODULE_VERSION_1",
+       "batch_size": "BATCH_SIZE_1"
+    },  
+    {  
+       "module": "MODULE_NAME_2",  
+       "version": "MODULE_VERSION_2",  
+       "batch_size": "BATCH_SIZE_2"
+    }
+  ],  
+  "use_gpu": false,  
+  "port": 8866,
+  "use_multiprocess": false
+}
+```
+
+**参数**：
+
+|参数|用途|  
+|-|-|  
+|--modules_info|Hub-Serving预安装模型，以字典列表形式列出，其中:<br>`module`为预测服务使用的模型名<br>`version`为预测模型的版本<br>`batch_size`为预测批次大小
+|--use_gpu|使用GPU进行预测，必须安装paddlepaddle-gpu|  
+|--port/-p|服务端口，默认为8866|  
+|--use_multiprocess|是否启用并发方式，默认为单进程方式，推荐多核CPU机器使用此方式|  
+
+### 2.2 访问PaddleHub-Serving服务端
+
+在使用PaddleHub-Serving部署服务端的模型预测服务后，就可以在客户端访问预测接口以获取结果了，接口url格式为：
+
+http://0.0.0.0:8866/predict/\<CATEGORY\>/\<MODULE>
+
+其中，\<CATEGORY>为text或image，与模型种类对应，\<MODULE>为模型名。
+
+通过发送一个POST请求，即可获取预测结果，下面我们将展示一个具体的demo，以说明使用PaddleHub-Serving部署和使用流程。
+
+### 2.3 利用PaddleHub-Serving进行个性化开发
+使用PaddleHub-Serving进行模型服务部署后，可以利用得到的接口进行开发，如对外提供web服务，或接入到应用程序中，以降低客户端预测压力，提高性能，下面展示了一个web页面demo:
+
+<p align="center">  
+
+<img src="./img/web_demo.png" width="60%" />  
+
+</p>  
+
+## 3 示例-部署一个在线lac分词服务
+
+### 3.1 部署lac在线服务
+现在，我们要部署一个lac在线服务，以通过接口获取文本的分词结果。
+
+首先，根据2.1节所述，启动PaddleHub-Serving服务端的两种方式分别为:
+```shell
+$ hub serving start -m lac
+```
+或
+```shell
+$ hub serving start -c serving_config.json
+```
+其中`serving_config.json`的内容如下：
+```json
+{
+  "modules_info": [
+  {
+    "module": "lac",
+    "version": "1.0.0",
+    "batch_size": 1
+  }
+],
+  "use_gpu": false,
+  "port": 8866,
+  "use_multiprocess": false
+}
+```
+启动成功界面如图：
+
+<p align="center">  
+
+<img src="./img/start_serving_lac.png" width="100%" />  
+
+</p>  
+
+这样我们就在8866端口部署了lac的在线分词服务。
+*此处warning为Flask提示，不影响使用*
+
+### 3.2 访问lac预测接口
+
+在服务部署好之后，我们可以进行测试，用来测试的文本为`今天是个好日子`和`天气预报说今天要下雨`。
+
+客户端代码如下
+```python
+# coding: utf8
+import requests
+import json
+
+if __name__ == "__main__":
+    # 指定用于用于预测的文本并生成字典{"text": [text_1, text_2, ... ]}
+    text_list = ["今天是个好日子", "天气预报说今天要下雨"]
+    text = {"text": text_list}
+    # 指定预测方法为lac并发送post请求
+    url = "http://127.0.0.1:8866/predict/text/lac"
+    r = requests.post(url=url, data=text)
+
+    # 打印预测结果
+    print(json.dumps(r.json(), indent=4, ensure_ascii=False))
+```
+运行后得到结果
+<details>
+    <summary>lac分词结果</summary>
+
+    ```python
+    {
+        "results": [
+            {
+                "tag": [
+                    "TIME",
+                    "v",
+                    "q",
+                    "n"
+                ],
+                "word": [
+                    "今天",
+                    "是",
+                    "个",
+                    "好日子"
+                ]
+            },
+            {
+                "tag": [
+                    "n",
+                    "v",
+                    "TIME",
+                    "v",
+                    "v"
+                ],
+                "word": [
+                    "天气预报",
+                    "说",
+                    "今天",
+                    "要",
+                    "下雨"
+                ]
+            }
+        ]
+    }
+    ```
+
+</detail>  
+
+获取其他PaddleHub-Serving的一键服务部署场景示例，可参阅下列demo
+
+* [图像分类-基于vgg11_imagent](serving/classification_vgg11_imagenet)  
+
+该示例展示了利用vgg11_imagent完成图像分类服务化部署和在线预测，获取图像分类结果。
+
+* [图像生成-基于stgan_celeba](serving/GAN_stgan_celeba)  
+
+该示例展示了利用stgan_celeba生成图像服务化部署和在线预测，获取指定风格的生成图像。
+
+* [文本审核-基于porn_detection_lstm](serving/text_censorship_porn_detection_lstm)  
+
+该示例展示了利用porn_detection_lstm完成中文文本黄色敏感信息鉴定的服务化部署和在线预测，获取文本是否敏感及其置信度。
+
+* [中文词法分析-基于lac](serving/lexical_analysis_lac)
+
+该示例展示了利用lac完成中文文本分词服务化部署和在线预测，获取文本的分词结果，并可通过用户自定义词典干预分词结果。
+
+* [目标检测-基于yolov3_darknet53_coco2017](serving/object_detection_yolov3_darknet53_coco2017)  
+
+该示例展示了利用yolov3_darknet53_coco2017完成目标检测服务化部署和在线预测，获取检测结果和覆盖识别框的图片。
+
+* [中文语义分析-基于simnet_bow](serving/semantic_model_simnet_bow)
+
+该示例展示了利用simnet_bow完成中文文本相似度检测服务化部署和在线预测，获取文本的相似程度。  
+
+* [图像分割-基于deeplabv3p_xception65_humanseg](serving/semantic_segmentation_deeplabv3p_xception65_humanseg)
+
+该示例展示了利用deeplabv3p_xception65_humanseg完成图像分割服务化部署和在线预测，获取识别结果和分割后的图像。
+
+* [中文情感分析-基于simnet_bow](serving/semantic_model_simnet_bow)
+
+该示例展示了利用senta_lstm完成中文文本情感分析服务化部署和在线预测，获取文本的情感分析结果。
diff --git a/demo/serving/serving/Sentiment_Analysis_senta_lstm/README.md b/demo/serving/serving/Sentiment_Analysis_senta_lstm/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..be187024b153b51b7992912b83bcead3d39c3f5c
--- /dev/null
+++ b/demo/serving/serving/Sentiment_Analysis_senta_lstm/README.md
@@ -0,0 +1,82 @@
+# 部署情感分析服务-以senta_lstm为例
+## 1 简介
+情感分析针对带有主观描述的中文文本，可自动判断该文本的情感极性类别并给出相应的置信度。利用`senta_lstm`模型可以完成中文情感分析任务，关于`senta_lstm`的具体信息请参阅[senta_lstm](https://paddlepaddle.org.cn/hubdetail?name=senta_lstm&en_category=SentimentAnalysis)。
+
+使用PaddleHub-Serving可以部署一个在线情感分析服务，可以将此接口用于分析评论、智能客服等应用。
+
+这里就带领大家使用PaddleHub-Serving，通过简单几步部署一个情感分析在线服务。
+
+## 2 启动PaddleHub-Serving
+启动命令如下
+```shell
+$ hub serving start -m senta_lstm  
+```
+启动时会显示加载模型过程，启动成功后显示
+```shell
+Loading senta_lstm successful.
+```
+这样就完成了一个词法分析服务化API的部署，默认端口号为8866。
+
+## 3 测试词法分析在线API
+在服务部署好之后，我们可以进行测试，用来测试的文本为`我不爱吃甜食`和`我喜欢躺在床上看电影`。
+
+准备的数据格式为
+```python
+{"text": [text_1, text_2, ...]}  
+```
+注意字典的key为"text"。
+
+根据文本和数据格式，代码如下
+```python
+>>> # 指定用于用于预测的文本并生成字典{"text": [text_1, text_2, ... ]}
+>>> text_list = ["我不爱吃甜食", "我喜欢躺在床上看电影"]
+>>> text = {"text": text_list}
+```
+接下来发送请求到词法分析API，并得到结果，代码如下
+```python
+# 指定预测方法为lac并发送post请求
+>>> url = "http://127.0.0.1:8866/predict/text/senta_lstm"
+>>> r = requests.post(url=url, data=text)
+```
+`LAC`模型返回的结果为每个文本分词后的结果，我们尝试打印接口返回结果
+```python
+# 打印预测结果
+>>> print(json.dumps(r.json(), indent=4, ensure_ascii=False))
+{
+    "results": [
+        {
+            "tag": [
+                "TIME",
+                "v",
+                "q",
+                "n"
+            ],
+            "word": [
+                "今天",
+                "是",
+                "个",
+                "好日子"
+            ]
+        },
+        {
+            "tag": [
+                "n",
+                "v",
+                "TIME",
+                "v",
+                "v"
+            ],
+            "word": [
+                "天气预报",
+                "说",
+                "今天",
+                "要",
+                "下雨"
+            ]
+        }
+    ]
+}
+```
+这样我们就完成了对词法分析的预测服务化部署和测试。
+
+完整的测试代码见[senta_lstm_serving_demo.py](senta_lstm_serving_demo.py)。
diff --git a/demo/serving/Sentiment_Analysis_senta_lstm/senta_lstm_serving_demo.py b/demo/serving/serving/Sentiment_Analysis_senta_lstm/senta_lstm_serving_demo.py
similarity index 100%
rename from demo/serving/Sentiment_Analysis_senta_lstm/senta_lstm_serving_demo.py
rename to demo/serving/serving/Sentiment_Analysis_senta_lstm/senta_lstm_serving_demo.py
diff --git a/demo/serving/serving/classification_vgg11_imagenet/README.md b/demo/serving/serving/classification_vgg11_imagenet/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..5dc9e974004279b270f8003d0eeda325c029d469
--- /dev/null
+++ b/demo/serving/serving/classification_vgg11_imagenet/README.md
@@ -0,0 +1,59 @@
+# 部署图像分类服务-以vgg11_imagenent为例
+## 1 简介
+图像分类是指通过模型，预测给定的图片所属类别，vgg11_imagenent就是一种有效的图像分类模型。关于vgg11_imagenent的具体信息请参阅[vgg11_imagenent](https://paddlepaddle.org.cn/hubdetail?name=vgg11_imagenet&en_category=ImageClassification)。
+
+使用PaddleHub-Serving可以部署一个在线图片分类服务，既可以对用户暴露直接预测接口，也可以利用此接口实现一个web网站，甚至可以集成到移动端应用程序中实现拍照识别功能。
+
+这里就带领大家使用PaddleHub-Serving，通过简单几步部署一个图像分类服务。
+
+## 2 启动PaddleHub-Serving
+启动命令如下
+```shell
+$ hub serving start -m vgg11_imagenet  
+```
+启动时会显示加载模型过程，启动成功后显示
+```shell
+Loading vgg11_imagenet successful.
+```
+这样就完成了一个图像分类服务化API的部署，默认端口号为8866。
+
+## 3 测试图像分类在线API
+我们用来测试的样例图片为  
+
+<p align="center">  
+<img src="../img/cat.jpg" width="45%" />  
+</p>  
+
+<p align="center">  
+<img src="../img/flower.jpg" width="45%"/>  
+</p>
+
+准备的数据格式为
+```python
+files = [("image", file_1), ("image", file_2)]
+```
+注意每个元素第一个参数为"image"。
+
+代码如下
+```python
+>>> file_list = ["../img/cat.jpg", "../img/flower.jpg"]  
+>>> files = [("image", (open(item, "rb"))) for item in file_list]
+```
+然后就可以发送请求到图像分类服务API，并得到结果了，代码如下
+```python
+>>> # 指定检测方法为vgg11_imagenet并发送post请求
+>>> url = "http://127.0.0.1:8866/predict/image/vgg11_imagenet"
+>>> r = requests.post(url=url, files=files)
+```
+vgg11_imagenent返回的结果为图像分类结果及其对应的概率，我们尝试打印接口返回结果
+```python
+
+>>> print(json.dumps(r.json(), indent=4, ensure_ascii=False))  
+{
+    "results": "[[{'Egyptian cat': 0.540287435054779}], [{'daisy': 0.9976677298545837}]]"
+}
+```
+
+这样我们就完成了对图像分类预测服务化部署和测试。
+
+完整的测试代码见[vgg11_imagenent_serving_demo.py](vgg11_imagenet_serving_demo.py)。
diff --git a/demo/serving/Classification_vgg11_imagenet/vgg11_imagenet_serving_demo.py b/demo/serving/serving/classification_vgg11_imagenet/vgg11_imagenet_serving_demo.py
similarity index 100%
rename from demo/serving/Classification_vgg11_imagenet/vgg11_imagenet_serving_demo.py
rename to demo/serving/serving/classification_vgg11_imagenet/vgg11_imagenet_serving_demo.py
diff --git a/demo/serving/img/cat.jpg b/demo/serving/serving/img/cat.jpg
similarity index 100%
rename from demo/serving/img/cat.jpg
rename to demo/serving/serving/img/cat.jpg
diff --git a/demo/serving/img/dog.jpg b/demo/serving/serving/img/dog.jpg
similarity index 100%
rename from demo/serving/img/dog.jpg
rename to demo/serving/serving/img/dog.jpg
diff --git a/demo/serving/img/flower.jpg b/demo/serving/serving/img/flower.jpg
similarity index 100%
rename from demo/serving/img/flower.jpg
rename to demo/serving/serving/img/flower.jpg
diff --git a/demo/serving/img/girl.jpg b/demo/serving/serving/img/girl.jpg
similarity index 100%
rename from demo/serving/img/girl.jpg
rename to demo/serving/serving/img/girl.jpg
diff --git a/demo/serving/serving/img/man.png b/demo/serving/serving/img/man.png
new file mode 100644
index 0000000000000000000000000000000000000000..2a6b9755d77b7a986a7e2144034772ace2af91ff
Binary files /dev/null and b/demo/serving/serving/img/man.png differ
diff --git a/demo/serving/serving/img/start_serving_lac.png b/demo/serving/serving/img/start_serving_lac.png
new file mode 100644
index 0000000000000000000000000000000000000000..67b1ae337549305c95bceb810fd448c3697e0dd5
Binary files /dev/null and b/demo/serving/serving/img/start_serving_lac.png differ
diff --git a/demo/serving/serving/img/web_demo.png b/demo/serving/serving/img/web_demo.png
new file mode 100644
index 0000000000000000000000000000000000000000..169f426251d243327bbbb05bf056cc27a7a1d7b5
Binary files /dev/null and b/demo/serving/serving/img/web_demo.png differ
diff --git a/demo/serving/img/woman.png b/demo/serving/serving/img/woman.png
similarity index 100%
rename from demo/serving/img/woman.png
rename to demo/serving/serving/img/woman.png
diff --git a/demo/serving/serving/lexical_analysis_lac/README.md b/demo/serving/serving/lexical_analysis_lac/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..8405f9e503595886f26c6ce347e558d2f4b96299
--- /dev/null
+++ b/demo/serving/serving/lexical_analysis_lac/README.md
@@ -0,0 +1,99 @@
+# 部署词法分析服务-以lac为例
+## 1 简介
+`Lexical Analysis of Chinese`，简称`LAC`，是一个联合的词法分析模型，能整体性地完成中文分词、词性标注、专名识别任务。关于`LAC`的具体信息请参阅[LAC](https://paddlepaddle.org.cn/hubdetail?name=lac&en_category=LexicalAnalysis)。
+
+使用PaddleHub-Serving可以部署一个在线词法分析服务，可以将此接口用于词法分析、在线分词等在线web应用。
+
+这里就带领大家使用PaddleHub-Serving，通过简单几步部署一个词法分析在线服务。
+
+## 2 启动PaddleHub-Serving
+启动命令如下
+```shell
+$ hub serving start -m lac  
+```
+启动时会显示加载模型过程，启动成功后显示
+```shell
+Loading lac successful.
+```
+这样就完成了一个词法分析服务化API的部署，默认端口号为8866。
+
+## 3 测试语言模型在线API
+### 3.1 不使用自定义词典
+在服务部署好之后，我们可以进行测试，用来测试的文本为`今天是个好日子`和`天气预报说今天要下雨`。
+
+准备的数据格式为
+```python
+{"text": [text_1, text_2, ...]}  
+```
+注意字典的key为"text"。
+
+根据文本和数据格式，代码如下
+```python
+>>> # 指定用于用于预测的文本并生成字典{"text": [text_1, text_2, ... ]}
+>>> text_list = ["今天是个好日子", "天气预报说今天要下雨"]
+>>> text = {"text": text_list}
+```
+接下来发送请求到词法分析API，并得到结果，代码如下
+```python
+# 指定预测方法为lac并发送post请求
+>>> url = "http://127.0.0.1:8866/predict/text/lac"
+>>> r = requests.post(url=url, data=text)
+```
+`LAC`模型返回的结果为每个文本分词后的结果，我们尝试打印接口返回结果
+```python
+# 打印预测结果
+>>> print(json.dumps(r.json(), indent=4, ensure_ascii=False))
+{
+    "results": [
+        {
+            "tag": [
+                "TIME",
+                "v",
+                "q",
+                "n"
+            ],
+            "word": [
+                "今天",
+                "是",
+                "个",
+                "好日子"
+            ]
+        },
+        {
+            "tag": [
+                "n",
+                "v",
+                "TIME",
+                "v",
+                "v"
+            ],
+            "word": [
+                "天气预报",
+                "说",
+                "今天",
+                "要",
+                "下雨"
+            ]
+        }
+    ]
+}
+```
+这样我们就完成了对词法分析的预测服务化部署和测试。
+
+完整的测试代码见[lac_serving_demo.py](lac_serving_demo.py)。
+
+### 3.2 使用自定义词典
+`LAC`模型在预测时还可以使用自定义词典干预默认分词结果，这种情况只需要将自定义词典以文件的形式附加到request请求即可，数据格式如下
+```python
+{"user_dict": user_dict.txt}
+```
+根据数据格式，具体代码如下
+```python
+>>> # 指定自定义词典{"user_dict": dict.txt}
+>>> file = {"user_dict": open("dict.txt", "rb")}
+>>> # 请求接口时以文件的形式附加自定义词典，其余和不使用自定义词典的请求方式相同，此处不再赘述
+>>> url = "http://127.0.0.1:8866/predict/text/lac"
+>>> r = requests.post(url=url, files=file, data=text)
+```
+
+完整的测试代码见[lac_with_dict_serving_demo.py](lac_with_dict_serving_demo.py)。
diff --git a/demo/serving/Lexical_Analysis_lac/dict.txt b/demo/serving/serving/lexical_analysis_lac/dict.txt
similarity index 100%
rename from demo/serving/Lexical_Analysis_lac/dict.txt
rename to demo/serving/serving/lexical_analysis_lac/dict.txt
diff --git a/demo/serving/Lexical_Analysis_lac/lac_serving_demo.py b/demo/serving/serving/lexical_analysis_lac/lac_serving_demo.py
similarity index 100%
rename from demo/serving/Lexical_Analysis_lac/lac_serving_demo.py
rename to demo/serving/serving/lexical_analysis_lac/lac_serving_demo.py
diff --git a/demo/serving/Lexical_Analysis_lac/lac_with_dict_serving_demo.py b/demo/serving/serving/lexical_analysis_lac/lac_with_dict_serving_demo.py
similarity index 100%
rename from demo/serving/Lexical_Analysis_lac/lac_with_dict_serving_demo.py
rename to demo/serving/serving/lexical_analysis_lac/lac_with_dict_serving_demo.py
diff --git a/demo/serving/serving/object_detection_yolov3_darknet53_coco2017/README.md b/demo/serving/serving/object_detection_yolov3_darknet53_coco2017/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..2129ed0c0e4116037d816d443038d2019b98031d
--- /dev/null
+++ b/demo/serving/serving/object_detection_yolov3_darknet53_coco2017/README.md
@@ -0,0 +1,119 @@
+# 部署图像分类服务-以yolov3_darknet53_coco2017为例
+## 1 简介
+目标检测作为深度学习常见任务，在各种场景下都有所使用。使用`yolov3_darknet53_coco2017`模型可以进行目标检测任务，关于`yolov3_darknet53_coco2017`的具体信息请参阅[yolov3_darknet53_coco2017](https://paddlepaddle.org.cn/hubdetail?name=yolov3_darknet53_coco2017&en_category=ObjectDetection)。
+
+使用PaddleHub-Serving可以轻松部署一个在线目标检测服务API，可将此API接入自己的web网站进行在线目标检测，也可接入移动端应用程序，实现识图、圈人等功能。
+
+下面就带领大家使用PaddleHub-Serving，通过简单几步部署一个目标检测服务。
+
+## 2 启动PaddleHub-Serving
+启动命令如下
+```shell
+$ hub serving start -m yolov3_darknet53_coco2017
+```
+启动时会显示加载模型过程，启动成功后显示
+```shell
+Loading yolov3_darknet53_coco2017 successful.
+```
+这样就完成了一个图像生成服务化API的部署，默认端口号为8866。
+
+## 3 测试图像生成在线API
+我们用来测试的样例图片为  
+
+<p align="center">  
+
+<img src="../img/cat.jpg" width="65%" />  
+
+</p>  
+
+<p align="center">  
+
+<img src="../img/dog.jpg" width="65%" />  
+
+</p>  
+
+准备的数据格式为
+```python
+files = [("image", file_1), ("image", file_2)]
+```
+注意文件列表每个元素第一个参数为"image"。
+
+代码如下
+```python
+>>> # 指定要检测的图片并生成列表[("image", img_1), ("image", img_2), ... ]
+>>> file_list = ["../img/cat.jpg", "../img/dog.jpg"]
+>>> files = [("image", (open(item, "rb"))) for item in file_list]
+```
+然后就可以发送请求到目标检测服务API，并得到结果，代码如下
+```python
+>>> # 指定检测方法为yolov3_darknet53_coco2017并发送post请求
+>>> url = "http://127.0.0.1:8866/predict/image/yolov3_darknet53_coco2017"
+>>> r = requests.post(url=url, files=files)
+```
+我们可以打印接口返回结果
+```python
+>>> results = eval(r.json()["results"])
+>>> print(json.dumps(results, indent=4, ensure_ascii=False))
+[
+    {
+        "path": "cat.jpg",
+        "data": [
+            {
+                "left": 319.489,
+                "right": 1422.8364,
+                "top": 208.94229,
+                "bottom": 993.8552,
+                "label": "cat",
+                "confidence": 0.9174191
+            }
+        ]
+    },
+    {
+        "path": "dog.jpg",
+        "data": [
+            {
+                "left": 200.6918,
+                "right": 748.96204,
+                "top": 122.74927,
+                "bottom": 566.2066,
+                "label": "dog",
+                "confidence": 0.83619183
+            },
+            {
+                "left": 506.8462,
+                "right": 623.2322,
+                "top": 378.0084,
+                "bottom": 416.116,
+                "label": "tie",
+                "confidence": 0.5082839
+            }
+        ]
+    }
+]
+```
+根据结果可以看出准确识别了请求的图片。
+
+yolov3_darknet53_coco2017返回的结果还包括标注检测框的图像的base64编码格式，经过转换可以得到生成图像，代码如下
+```python
+>>> for item in results:
+...     with open(output_path, "wb") as fp:
+...         fp.write(base64.b64decode(item["base64"].split(',')[-1]))
+```
+查看指定输出文件夹，就能看到生成图像了，如图
+
+<p align="center">  
+
+<img src="./output/cat.jpg" width="65%" />  
+
+</p>  
+
+<p align="center">  
+
+<img src="./output/dog.jpg" width="65%" />  
+
+</p>  
+
+
+这样我们就完成了对目标检测服务化的部署和测试。
+
+完整的测试代码见[yolov3_darknet53_coco2017_serving_demo.py](yolov3_darknet53_coco2017_serving_demo.py)。
diff --git a/demo/serving/serving/object_detection_yolov3_darknet53_coco2017/output/cat.jpg b/demo/serving/serving/object_detection_yolov3_darknet53_coco2017/output/cat.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..d71260a8c364349bc0f07ef9938220eef893d857
Binary files /dev/null and b/demo/serving/serving/object_detection_yolov3_darknet53_coco2017/output/cat.jpg differ
diff --git a/demo/serving/serving/object_detection_yolov3_darknet53_coco2017/output/dog.jpg b/demo/serving/serving/object_detection_yolov3_darknet53_coco2017/output/dog.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..b29ca3439b1e1f462ebf2a97ba732a7826ac834f
Binary files /dev/null and b/demo/serving/serving/object_detection_yolov3_darknet53_coco2017/output/dog.jpg differ
diff --git a/demo/serving/Object_Detection_yolov3_coco2017/yolov3_coco2017_serving_demo.py b/demo/serving/serving/object_detection_yolov3_darknet53_coco2017/yolov3_darknet53_coco2017_serving_demo.py
similarity index 91%
rename from demo/serving/Object_Detection_yolov3_coco2017/yolov3_coco2017_serving_demo.py
rename to demo/serving/serving/object_detection_yolov3_darknet53_coco2017/yolov3_darknet53_coco2017_serving_demo.py
index e8f9409370a9cd550bb76313978ffd787c71c436..8b85f223ada1116ddefa8000fadb42b55eb02369 100644
--- a/demo/serving/Object_Detection_yolov3_coco2017/yolov3_coco2017_serving_demo.py
+++ b/demo/serving/serving/object_detection_yolov3_darknet53_coco2017/yolov3_darknet53_coco2017_serving_demo.py
@@ -9,7 +9,7 @@ if __name__ == "__main__":
     file_list = ["../img/cat.jpg", "../img/dog.jpg"]
     files = [("image", (open(item, "rb"))) for item in file_list]
     # 指定检测方法为yolov3_coco2017并发送post请求
-    url = "http://127.0.0.1:8866/predict/image/yolov3_coco2017"
+    url = "http://127.0.0.1:8866/predict/image/yolov3_darknet53_coco2017"
     r = requests.post(url=url, files=files)
 
     results = eval(r.json()["results"])
diff --git a/demo/serving/serving/semantic_model_simnet_bow/README.md b/demo/serving/serving/semantic_model_simnet_bow/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..05e72cf41e91586758f968e608ae2d7fd5ee549f
--- /dev/null
+++ b/demo/serving/serving/semantic_model_simnet_bow/README.md
@@ -0,0 +1,70 @@
+# 部署语义模型服务-以simnet_bow为例
+## 1 简介
+`simnet_bow`是一个计算短文本相似度的模型，可以根据用户输入的两个文本，计算出相似度得分。关于`simnet_bow`的具体信息请参阅[simnet_bow](https://paddlepaddle.org.cn/hubdetail?name=simnet_bow&en_category=SemanticModel)。
+
+使用PaddleHub-Serving可以部署一个在线语义模型服务，可以将此接口用于在线文本相似度分析、智能问答检索等应用。
+
+这里就带领大家使用PaddleHub-Serving，通过简单几步部署一个语义模型在线服务。
+
+## 2 启动PaddleHub-Serving
+启动命令如下
+```shell
+$ hub serving start -m simnet_bow  
+```
+启动时会显示加载模型过程，启动成功后显示
+```shell
+Loading lac successful.
+```
+这样就完成了一个语义模型服务化API的部署，默认端口号为8866。
+
+## 3 测试语义模型在线API
+在服务部署好之后，我们可以进行测试，用来测试的文本对分别为`[这道题太难了:这道题是上一年的考题], [这道题太难了:这道题不简单], [这道题太难了:这道题很有意思]`。
+
+准备的数据格式为
+```python
+{"text_1": [text_a1, text_a2, ... ], "text_2": [text_b1, text_b2, ... ]}
+```
+注意字典的key分别为"text_1"和"text_2"，与`simnet_bow`模型使用的输入数据一致。
+
+根据文本和数据格式，代码如下
+```python
+>>> # 指定用于用于匹配的文本并生成字典{"text_1": [text_a1, text_a2, ... ]
+>>> #                              "text_2": [text_b1, text_b2, ... ]}
+>>> text = {
+>>>     "text_1": ["这道题太难了", "这道题太难了", "这道题太难了"],
+>>>     "text_2": ["这道题是上一年的考题", "这道题不简单", "这道题很有意思"]
+>>> }
+```
+接下来发送请求到语义模型API，并得到结果，代码如下
+```python
+>>> # 指定匹配方法为simnet_bow并发送post请求
+>>> url = "http://127.0.0.1:8866/predict/text/simnet_bow"
+>>> r = requests.post(url=url, data=text)
+```
+`simnet_bow`模型返回的结果为每对文本对比后的相似度，我们尝试打印接口返回结果
+```python
+# 打印预测结果
+>>> print(json.dumps(r.json(), indent=4, ensure_ascii=False))
+{
+    "results": [
+        {
+            "similarity": 0.8445,
+            "text_1": "这道题太难了",
+            "text_2": "这道题是上一年的考题"
+        },
+        {
+            "similarity": 0.9275,
+            "text_1": "这道题太难了",
+            "text_2": "这道题不简单"
+        },
+        {
+            "similarity": 0.9083,
+            "text_1": "这道题太难了",
+            "text_2": "这道题很有意思"
+        }
+    ]
+}
+```
+这样我们就完成了对语义模型simnet_bow的预测服务化部署和测试。
+
+完整的测试代码见[simnet_bow_serving_demo.py](simnet_bow_serving_demo.py)。
diff --git a/demo/serving/Semantic_Model_simnet_bow/simnet_bow_serving_demo.py b/demo/serving/serving/semantic_model_simnet_bow/simnet_bow_serving_demo.py
similarity index 100%
rename from demo/serving/Semantic_Model_simnet_bow/simnet_bow_serving_demo.py
rename to demo/serving/serving/semantic_model_simnet_bow/simnet_bow_serving_demo.py
diff --git a/demo/serving/serving/semantic_segmentation_deeplabv3p_xception65_humanseg/README.md b/demo/serving/serving/semantic_segmentation_deeplabv3p_xception65_humanseg/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..3a1a11535c8b069329c2b1268d917fc867d7b785
--- /dev/null
+++ b/demo/serving/serving/semantic_segmentation_deeplabv3p_xception65_humanseg/README.md
@@ -0,0 +1,75 @@
+# 部署图像分割服务-以deeplabv3p_xception65_humanseg为例
+## 1 简介
+图像分割是深度学习的常见任务。使用`deeplabv3p_xception65_humanseg`模型可以进行人像分割任务，关于`deeplabv3p_xception65_humanseg`的具体信息请参阅[deeplabv3p_xception65_humanseg](https://paddlepaddle.org.cn/hubdetail?name=deeplabv3p_xception65_humanseg&en_category=ImageSegmentation)。
+
+使用PaddleHub-Serving可以轻松部署一个在线图像分割服务API，可将此API接入自己的web网站进行在线图像分割，也可接入移动端应用程序，实现拍照分割等功能。
+
+下面就带领大家使用PaddleHub-Serving，通过简单几步部署一个目标检测服务。
+
+## 2 启动PaddleHub-Serving
+启动命令如下
+```shell
+$ hub serving start -m deeplabv3p_xception65_humanseg
+```
+启动时会显示加载模型过程，启动成功后显示
+```shell
+Loading deeplabv3p_xception65_humanseg successful.
+```
+这样就完成了一个图像分割服务化API的部署，默认端口号为8866。
+
+## 3 测试图像分割在线API
+我们用来测试的样例图片为  
+
+<p align="center">  
+
+<img src="../img/girl.jpg" width="65%" />  
+
+</p>  
+
+准备的数据格式为
+```python
+files = [("image", file_1), ("image", file_2)]
+```
+注意文件列表每个元素第一个参数为"image"。
+
+代码如下
+```python
+>>> # 指定要检测的图片并生成列表[("image", img_1), ("image", img_2), ... ]
+>>> file_list = ["../img/girl.jpg"]
+>>> files = [("image", (open(item, "rb"))) for item in file_list]
+```
+然后就可以发送请求到图像分割服务API，并得到结果，代码如下
+```python
+>>> # 指定检测方法为deeplabv3p_xception65_humanseg并发送post请求
+>>> url = "http://127.0.0.1:8866/predict/image/deeplabv3p_xception65_humanseg"
+>>> r = requests.post(url=url, files=files)
+```
+我们可以打印接口返回结果
+```python
+>>> results = eval(r.json()["results"])
+>>> print(json.dumps(results, indent=4, ensure_ascii=False))
+[
+    {
+        "origin": "girl.jpg",
+        "processed": "humanseg_output/girl.png"
+    }
+]
+```
+
+deeplabv3p_xception65_humanseg返回的结果还包括人像分割后的图像的base64编码格式，经过转换可以得到生成图像，代码如下
+```python
+>>> for item in results:
+...     with open(output_path, "wb") as fp:
+...         fp.write(base64.b64decode(item["base64"].split(',')[-1]))
+```
+查看指定输出文件夹，就能看到生成图像了，如图
+
+<p align="center">  
+
+<img src="./output/girl.png" width="65%" />  
+
+</p>  
+
+这样我们就完成了对图像分割模型deeplabv3p_xception65_humanseg服务化的部署和测试。
+
+完整的测试代码见[deeplabv3p_xception65_humanseg_serving_demo.py](deeplabv3p_xception65_humanseg_serving_demo.py)。
diff --git a/demo/serving/Semantic_Segmentation_deeplabv3p_xception65_humanseg/deeplabv3p_xception65_humanseg_serving_demo.py b/demo/serving/serving/semantic_segmentation_deeplabv3p_xception65_humanseg/deeplabv3p_xception65_humanseg_serving_demo.py
similarity index 100%
rename from demo/serving/Semantic_Segmentation_deeplabv3p_xception65_humanseg/deeplabv3p_xception65_humanseg_serving_demo.py
rename to demo/serving/serving/semantic_segmentation_deeplabv3p_xception65_humanseg/deeplabv3p_xception65_humanseg_serving_demo.py
diff --git a/demo/serving/serving/semantic_segmentation_deeplabv3p_xception65_humanseg/output/girl.png b/demo/serving/serving/semantic_segmentation_deeplabv3p_xception65_humanseg/output/girl.png
new file mode 100644
index 0000000000000000000000000000000000000000..667e9e70e5e8f0cf84cc5db7728eada13bf14607
Binary files /dev/null and b/demo/serving/serving/semantic_segmentation_deeplabv3p_xception65_humanseg/output/girl.png differ
diff --git a/demo/serving/serving/text_censorship_porn_detection_lstm/README.md b/demo/serving/serving/text_censorship_porn_detection_lstm/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..fcdbb36ad61972fa68762b071da3455cf6fe392a
--- /dev/null
+++ b/demo/serving/serving/text_censorship_porn_detection_lstm/README.md
@@ -0,0 +1,68 @@
+# 部署文本审核服务-以porn_detection_lstm为例
+## 1 简介
+在网站建设等场景中经常需要对敏感信息进行鉴定和过滤，采用文本审核模型`porn_detection_lstm`可自动判别文本是否涉黄并给出相应的置信度，关于`porn_detection_lstm`的具体信息请参阅[porn_detection_lstm](https://paddlepaddle.org.cn/hubdetail?name=porn_detection_lstm&en_category=TextCensorship)
+
+使用PaddleHub-Serving可以部署一个在线文本审核服务，可以将此接口用于防止低俗交友、色情文本等应用。
+
+这里就带领大家使用PaddleHub-Serving，通过简单几步部署一个文本审核在线服务。
+
+## 2 启动PaddleHub-Serving
+启动命令如下
+```shell
+$ hub serving start -m porn_detection_lstm  
+```
+启动时会显示加载模型过程，启动成功后显示
+```shell
+Loading porn_detection_lstm successful.
+```
+这样就完成了一个文本审核服务化API的部署，默认端口号为8866。
+
+## 3 测试文本审核在线API
+在服务部署好之后，我们可以进行测试，用来测试的文本为`黄片下载`和`中国黄页`。
+
+准备的数据格式为
+```python
+{"text": [text_1, text_2, ...]}  
+```
+注意字典的key为"text"。
+
+根据文本和数据格式，代码如下
+```python
+>>> # 指定用于用于预测的文本并生成字典{"text": [text_1, text_2, ... ]}
+>>> text_list = ["黄片下载", "中国黄页"]
+>>> text = {"text": text_list}
+```
+接下来发送请求到文本审核API，并得到结果，代码如下
+```python
+# 指定预测方法为lac并发送post请求
+>>> url = "http://127.0.0.1:8866/predict/text/porn_detection_lstm"
+>>> r = requests.post(url=url, data=text)
+```
+`porn_detection_lstm`模型返回的结果为每个文本鉴定后的结果，我们尝试打印接口返回结果
+```python
+# 打印预测结果
+>>> print(json.dumps(r.json(), indent=4, ensure_ascii=False))
+{
+    "results": [
+        {
+            "not_porn_probs": 0.0121,
+            "porn_detection_key": "porn",
+            "porn_detection_label": 1,
+            "porn_probs": 0.9879,
+            "text": "黄片下载"
+        },
+        {
+            "not_porn_probs": 0.9954,
+            "porn_detection_key": "not_porn",
+            "porn_detection_label": 0,
+            "porn_probs": 0.0046,
+            "text": "中国黄页"
+        }
+    ]
+}
+```
+可以看出正确得到了两个文本的预测结果。
+
+这样我们就完成了对文本审核模型的预测服务化部署和测试。
+
+完整的测试代码见[porn_detection_lstm_serving_demo.py](porn_detection_lstm_serving_demo.py)。
diff --git a/demo/serving/serving/text_censorship_porn_detection_lstm/porn_detection_lstm_serving_demo.py b/demo/serving/serving/text_censorship_porn_detection_lstm/porn_detection_lstm_serving_demo.py
new file mode 100644
index 0000000000000000000000000000000000000000..268a8d69a1c390ee6e8bd3874298cdaf101e6ad8
--- /dev/null
+++ b/demo/serving/serving/text_censorship_porn_detection_lstm/porn_detection_lstm_serving_demo.py
@@ -0,0 +1,14 @@
+# coding: utf8
+import requests
+import json
+
+if __name__ == "__main__":
+    # 指定用于用于预测的文本并生成字典{"text": [text_1, text_2, ... ]}
+    text_list = ["黄片下载", "中国黄页"]
+    text = {"text": text_list}
+    # 指定预测方法为lac并发送post请求
+    url = "http://127.0.0.1:8866/predict/text/porn_detection_lstm"
+    r = requests.post(url=url, data=text)
+
+    # 打印预测结果
+    print(json.dumps(r.json(), indent=4, ensure_ascii=False))
diff --git a/paddlehub/serving/app_single.py b/paddlehub/serving/app_single.py
index 24a44dab4f9687cca09b1f0e95eae125bae0e058..96fd56f6cd5fcf45a22239c690e309b74b87608f 100644
--- a/paddlehub/serving/app_single.py
+++ b/paddlehub/serving/app_single.py
@@ -133,7 +133,7 @@ def predict_gan(module, input_img, id, batch_size, extra={}):
 
 
 def predict_object_detection(module, input_img, id, batch_size, extra={}):
-    output_folder = "output"
+    output_folder = "detection_result"
     global use_gpu
     method_name = module.desc.attr.map.data['default_signature'].s
     predict_method = getattr(module, method_name)
diff --git a/paddlehub/serving/templates/main.html b/paddlehub/serving/templates/main.html
index 56208ea1ab83f14005ef09e613c9ad357f9e57b0..d272aae7f642fa359deddddf298ab2f1a57b3d76 100644
--- a/paddlehub/serving/templates/main.html
+++ b/paddlehub/serving/templates/main.html
@@ -115,7 +115,7 @@
                         + 'id="file_text"'
                         + 'onblur="blur_input_text()"'
                         + 'onfocus="focus_input_text()"'
-                        + 'name="input_text">'
+                        + 'name="text">'
                         + '在此键入文本或上传文本文件'
                     + '</textarea>'
                 + '</td>'
@@ -237,14 +237,15 @@
                     alert("Connection error:"+request.error);
                 },
                 success: function(data) {
-                    html = get_result_html(data["result"]);
+                    console.log(data);
+                    html = get_result_html(data["results"]);
                     document.getElementById("result_text").value = html;
                 }
             });
     }
     function sub_img() {
         var formParam = {
-            "input_img": document.getElementById("file_img").src,
+            "image": document.getElementById("file_img").src,
             "input_file":document.getElementById("file").value
         };
         to_url = "/predict/image/" + document.getElementById("inputGroupSelect02").value;
@@ -258,9 +259,13 @@
                     alert("Connection error:"+request.error);
                 },
                 success: function(data) {
-                    data = data["result"];
-                    document.getElementById("result_text").value = data["desc"];
-                    document.getElementById("result_img").src = data["output_img"];
+                    data = data.results;
+                    data = data.replace(/'/g, '"');
+                    data = JSON.parse(data);
+                    data = data[0];
+
+                    document.getElementById("result_text").value = JSON.stringify(data.data[0]);
+                    document.getElementById("result_img").src = data.base64;
                 }
             });
     }
diff --git a/paddlehub/serving/templates/serving_config.json b/paddlehub/serving/templates/serving_config.json
index 1d4196dd7ee2cebec1b0d1daf96b14be1454a435..cd761decf67696a052f86e2a7ba2359ef6264fe5 100644
--- a/paddlehub/serving/templates/serving_config.json
+++ b/paddlehub/serving/templates/serving_config.json
@@ -3,30 +3,22 @@
   {
     "module": "lac",
     "version": "1.0.0",
-    "batch_size": 200,
-    "queue_size": 200,
-    "category": "NLP"
+    "batch_size": 200
   },
   {
     "module": "senta_lstm",
     "version": "1.0.0",
-    "batch_size": 1,
-    "queue_size": 200,
-    "category": "NLP"
+    "batch_size": 1
   },
   {
     "module": "yolov3_darknet53_coco2017",
     "version": "1.0.0",
-    "batch_size": 1,
-    "queue_size": 10,
-    "category": "CV"
+    "batch_size": 1
   },
   {
     "module": "faster_rcnn_coco2017",
     "version": "1.0.0",
-    "batch_size": 1,
-    "queue_size": 10,
-    "category": "CV"
+    "batch_size": 1
   }
 ],
   "use_gpu": false,