diff --git a/deploy/pipeline/README.md b/deploy/pipeline/README.md
index cbb2d86fbe0a985545f0c466886611a9c4eca729..19534194700f4dcc270e63840302303d784e508b 100644
--- a/deploy/pipeline/README.md
+++ b/deploy/pipeline/README.md
@@ -128,7 +128,7 @@
 
 - [快速开始](docs/tutorials/ppvehicle_attribute.md)
 
-- 二次开发教程
+- [二次开发教程](../../docs/advanced_tutorials/customization/ppvehicle_attribute.md)
 
 #### 违章检测
 
diff --git a/docs/advanced_tutorials/customization/pphuman_attribute.md b/docs/advanced_tutorials/customization/pphuman_attribute.md
index 692a9dba57dba058db970995f2874732bab8e07d..75cdd4aebc22ddaae052c8f9be64e527d9f62fea 100644
--- a/docs/advanced_tutorials/customization/pphuman_attribute.md
+++ b/docs/advanced_tutorials/customization/pphuman_attribute.md
@@ -1,6 +1,6 @@
 简体中文 | [English](./pphuman_attribute_en.md)
 
-# 属性识别任务二次开发
+# 行人属性识别任务二次开发
 
 ## 数据准备
 
diff --git a/docs/advanced_tutorials/customization/pphuman_attribute_en.md b/docs/advanced_tutorials/customization/pphuman_attribute_en.md
index fdfe74eb2b3c531a0e03cc6e2c7e8ad78114f0b3..1dd3813ec3ff4c5e9ce66a7952b773e49555e307 100644
--- a/docs/advanced_tutorials/customization/pphuman_attribute_en.md
+++ b/docs/advanced_tutorials/customization/pphuman_attribute_en.md
@@ -1,6 +1,6 @@
 [简体中文](pphuman_attribute.md) | English
 
-# Customized attribute recognition
+# Customized Pedestrian Attribute Recognition
 
 ## Data Preparation
 
@@ -51,7 +51,7 @@ After the annotation is completed, the model will use the detection box to inter
 
 
 
-Model Training
+## Model Training
 
 Once the data is annotated, it can be used for model training to complete the optimization of the customized model.
 
diff --git a/docs/advanced_tutorials/customization/ppvehicle_attribute.md b/docs/advanced_tutorials/customization/ppvehicle_attribute.md
new file mode 100644
index 0000000000000000000000000000000000000000..c40c9cb3059ed9834fff89af7a5a3e78cb5a4bec
--- /dev/null
+++ b/docs/advanced_tutorials/customization/ppvehicle_attribute.md
@@ -0,0 +1,257 @@
+简体中文 | [English](./ppvehicle_attribute_en.md)
+
+# 车辆属性识别任务二次开发
+
+## 数据准备
+
+### 数据格式
+
+车辆属性模型采用VeRi数据集的属性，共计10种车辆颜色及9种车型, 具体如下：
+```
+# 车辆颜色
+- "yellow"
+- "orange"
+- "green"
+- "gray"
+- "red"
+- "blue"
+- "white"
+- "golden"
+- "brown"
+- "black"
+
+# 车型
+- "sedan"
+- "suv"
+- "van"
+- "hatchback"
+- "mpv"
+- "pickup"
+- "bus"
+- "truck"
+- "estate"
+```
+
+在标注文件中使用长度为19的序列来表示上述属性。
+
+举例：
+
+[1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0]
+
+前10位中，位序号0的值为1，表示车辆颜色为`"yellow"`。
+
+后9位中，位序号11的值为1，表示车型为`"suv"`。
+
+
+### 数据标注
+
+理解了上面`数据格式`的含义后，就可以进行数据标注的工作。其本质是：每张车辆的图片，建立一组长度为19的标注项，分别对应各项属性值。
+
+举例：
+
+对于一张原始图片，
+
+1） 使用检测框，标注图片中每台车辆的位置。
+
+2） 每一个检测框（对应每辆车），包含一组19位的属性值数组，数组的每一位以0或1表示。对应上述19个属性分类。例如，如果颜色是'orange'，则数组索引为1的位置值为1，如果车型是'sedan'，则数组索引为10的位置值为1。
+
+标注完成后利用检测框将每辆车截取成只包含单辆车的图片，则图片与19位属性标注建立了对应关系。也可先截取再进行标注，效果相同。
+
+
+## 模型训练
+
+数据标注完成后，就可以拿来做模型的训练，完成自定义模型的优化工作。
+
+其主要有两步工作需要完成：1）将数据与标注数据整理成训练格式。2）修改配置文件开始训练。
+
+### 训练数据格式
+
+训练数据包括训练使用的图片和一个训练列表train.txt，其具体位置在训练配置中指定，其放置方式示例如下：
+```
+Attribute/
+|-- data           训练图片文件夹
+|   |-- 00001.jpg
+|   |-- 00002.jpg
+|   `-- 0000x.jpg
+`-- train.txt      训练数据列表
+
+```
+
+train.txt文件内为所有训练图片名称（相对于根路径的文件路径）+ 19个标注值
+
+其每一行表示一辆车的图片和标注结果。其格式为：
+
+```
+00001.jpg    0,0,1,0,....
+```
+
+注意：1)图片与标注值之间是以Tab[\t]符号隔开, 2)标注值之间是以逗号[,]隔开。该格式不能错，否则解析失败。
+
+### 修改配置开始训练
+
+首先执行以下命令下载训练代码（更多环境问题请参考[Install_PaddleClas](https://github.com/PaddlePaddle/PaddleClas/blob/release/2.4/docs/en/installation/install_paddleclas_en.md)）:
+
+```shell
+git clone https://github.com/PaddlePaddle/PaddleClas
+```
+
+需要在[配置文件](https://github.com/PaddlePaddle/PaddleClas/blob/release/2.4/ppcls/configs/PULC/vehicle_attribute/PPLCNet_x1_0.yaml)中，修改的配置项如下：
+```yaml
+DataLoader:
+  Train:
+    dataset:
+      name: MultiLabelDataset
+      image_root: "dataset/VeRi/"                     # the root path of training images
+      cls_label_path: "dataset/VeRi/train_list.txt"   # the location of the training list file
+      label_ratio: True
+      transform_ops:
+        ...
+
+  Eval:
+    dataset:
+      name: MultiLabelDataset
+      image_root: "dataset/VeRi/"                     # the root path of evaluation images
+      cls_label_path: "dataset/VeRi/val_list.txt"     # the location of the evaluation list file
+      label_ratio: True
+      transform_ops:
+         ...
+```
+
+注意：
+1. 这里image_root路径+train.txt中图片相对路径，对应图片的完整路径位置。
+2. 如果有修改属性数量，则还需修改内容配置项中属性种类数量：
+```yaml
+# model architecture
+Arch:
+  name: "PPLCNet_x1_0"
+  pretrained: True
+  use_ssld: True
+  class_num: 19           #属性种类数量
+```
+
+然后运行以下命令开始训练。
+
+```
+#多卡训练
+export CUDA_VISIBLE_DEVICES=0,1,2,3
+python3 -m paddle.distributed.launch \
+    --gpus="0,1,2,3" \
+    tools/train.py \
+        -c ./ppcls/configs/PULC/vehicle_attribute/PPLCNet_x1_0.yaml
+
+#单卡训练
+python3 tools/train.py \
+        -c ./ppcls/configs/PULC/vehicle_attribute/PPLCNet_x1_0.yaml
+```
+
+训练完成后可以执行以下命令进行性能评估：
+```
+#多卡评估
+export CUDA_VISIBLE_DEVICES=0,1,2,3
+python3 -m paddle.distributed.launch \
+    --gpus="0,1,2,3" \
+    tools/eval.py \
+        -c ./ppcls/configs/PULC/vehicle_attribute/PPLCNet_x1_0.yaml \
+        -o Global.pretrained_model=./output/PPLCNet_x1_0/best_model
+
+#单卡评估
+python3 tools/eval.py \
+        -c ./ppcls/configs/PULC/vehicle_attribute/PPLCNet_x1_0.yaml \
+        -o Global.pretrained_model=./output/PPLCNet_x1_0/best_model
+```
+
+
+### 模型导出
+
+使用下述命令将训练好的模型导出为预测部署模型。
+
+```
+python3 tools/export_model.py \
+    -c ./ppcls/configs/PULC/vehicle_attribute/PPLCNet_x1_0.yaml \
+    -o Global.pretrained_model=output/PPLCNet_x1_0/best_model \
+    -o Global.save_inference_dir=deploy/models/PPLCNet_x1_0_vehicle_attribute_model
+```
+
+导出模型后，如果希望在PP-Vehicle中使用，则需要下载[预测部署模型](https://bj.bcebos.com/v1/paddledet/models/pipeline/vehicle_attribute_model.zip)，解压并将其中的配置文件`infer_cfg.yml`文件，放置到导出的模型文件夹`PPLCNet_x1_0_vehicle_attribute_model`中。
+
+使用时在PP-Vehicle中的配置文件`./deploy/pipeline/config/infer_cfg_ppvehicle.yml`中修改新的模型路径`model_dir`项，并开启功能`enable: True`。
+```
+VEHICLE_ATTR:
+  model_dir: [YOUR_DEPLOY_MODEL_DIR]/PPLCNet_x1_0_vehicle_attribute_infer/   #新导出的模型路径位置
+  enable: True                                                              #开启功能
+```
+然后可以使用-->至此即完成新增属性类别识别任务。
+
+## 属性增减
+
+该过程与行人属性的增减过程相似，如果需要增加、减少属性数量，则需要：
+
+1)标注时需增加新属性类别信息或删减属性类别信息；
+
+2)对应修改训练中train.txt所使用的属性数量和名称；
+
+3)修改训练配置，例如``PaddleClas/blob/develop/ppcls/configs/PULC/vehicle_attribute/PPLCNet_x1_0.yaml``文件中的属性数量,详细见上述`修改配置开始训练`部分。
+
+增加属性示例：
+
+1. 在标注数据时在19位后继续增加新的属性标注数值；
+2. 在train.txt文件的标注数值中也增加新的属性数值。
+3. 注意属性类型在train.txt中属性数值列表中的位置的对应关系需要固定。
+
+<div width="500" align="center">
+  <img src="../../images/add_attribute.png"/>
+</div>
+
+删减属性同理。
+
+
+## 修改后处理代码
+
+修改了属性定义后，pipeline后处理部分也需要做相应修改，主要影响结果可视化时的显示结果。
+
+相应代码在[文件](https://github.com/PaddlePaddle/PaddleDetection/blob/develop/deploy/pipeline/ppvehicle/vehicle_attr.py#L108)中`postprocess`函数。
+
+其函数实现说明如下：
+
+```python
+    # 在类的初始化函数中，定义了颜色/车型的名称
+    self.color_list = [
+        "yellow", "orange", "green", "gray", "red", "blue", "white",
+        "golden", "brown", "black"
+    ]
+    self.type_list = [
+        "sedan", "suv", "van", "hatchback", "mpv", "pickup", "bus", "truck",
+        "estate"
+    ]
+
+    ...
+
+    def postprocess(self, inputs, result):
+        # postprocess output of predictor
+        im_results = result['output']
+        batch_res = []
+        for res in im_results:
+            res = res.tolist()
+            attr_res = []
+            color_res_str = "Color: "
+            type_res_str = "Type: "
+            color_idx = np.argmax(res[:10])   # 前10项表示各项颜色得分，取得分最大项作为颜色结果
+            type_idx = np.argmax(res[10:])    # 后9项表示各项车型得分，取得分最大项作为车型结果
+
+            # 颜色和车型的得分都需要超过对应阈值，否则视为'UnKnown'
+            if res[color_idx] >= self.color_threshold:
+                color_res_str += self.color_list[color_idx]
+            else:
+                color_res_str += "Unknown"
+            attr_res.append(color_res_str)
+
+            if res[type_idx + 10] >= self.type_threshold:
+                type_res_str += self.type_list[type_idx]
+            else:
+                type_res_str += "Unknown"
+            attr_res.append(type_res_str)
+
+            batch_res.append(attr_res)
+        result = {'output': batch_res}
+        return result
+```
diff --git a/docs/advanced_tutorials/customization/ppvehicle_attribute_en.md b/docs/advanced_tutorials/customization/ppvehicle_attribute_en.md
new file mode 100644
index 0000000000000000000000000000000000000000..69fbc18fbf55081311978008d2b82c50f6f2634a
--- /dev/null
+++ b/docs/advanced_tutorials/customization/ppvehicle_attribute_en.md
@@ -0,0 +1,271 @@
+[简体中文](ppvehicle_attribute.md) | English
+
+# Customized Vehicle Attribute Recognition
+
+## Data Preparation
+
+### Data Format
+
+We use the VeRi attribute annotation format, with a total of 10 color and 9 model attributes shown as follows.
+
+```
+# colors
+- "yellow"
+- "orange"
+- "green"
+- "gray"
+- "red"
+- "blue"
+- "white"
+- "golden"
+- "brown"
+- "black"
+
+# models
+- "sedan"
+- "suv"
+- "van"
+- "hatchback"
+- "mpv"
+- "pickup"
+- "bus"
+- "truck"
+- "estate"
+```
+
+A sequence of length 19 is used in the annotation file to represent the above attributes.
+
+Examples:
+
+[1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0]
+
+In the first 10 bits, the value of bit index 0 is 1, indicating that the vehicle color is `"yellow"`.
+
+In the last 9 bits, the value of bit index 11 is 1, indicating that the model is `"suv"`.
+
+
+### Data Annotation
+
+After knowing the purpose of the above `Data format`, we can start to annotate data. The essence is that each single-vehicle image creates a set of 19 annotation items, corresponding to the attribute values at 19 positions.
+
+Examples:
+
+For an original image:
+
+1) Using bounding boxes to annotate the position of each vehicle in the picture.
+
+2) Each detection box (corresponding to each vehicle) contains 19 attribute values which are represented by 0 or 1. It corresponds to the above 19 attributes. For example, if the color is 'orange', then the index 1 bit of the array is 1. If the model is 'sedan', then the index 10 bit of the array is 1.
+
+After the annotation is completed, the model will use the detection box to intercept each vehicle into a single-vehicle picture, and its picture establishes a corresponding relationship with the 19 attribute annotation. It is also possible to cut into a single-vehicle image first and then annotate it. The results are the same.
+
+
+
+## Model Training
+
+Once the data is annotated, it can be used for model training to complete the optimization of the customized model.
+
+There are two main steps: 1) Organize the data and annotated data into the training format. 2) Modify the configuration file to start training.
+
+### Training Data Format
+
+The training data includes the images used for training and a training list called train.txt. Its location is specified in the training configuration, with the following example:
+
+```
+Attribute/
+|-- data      Training images folder
+|-- 00001.jpg
+|-- 00002.jpg
+| `-- 0000x.jpg
+train.txt     List of training data
+```
+
+train.txt file contains the names of all training images (file path relative to the root path) + 19 annotation values
+
+Each line of it represents a vehicle's image and annotation result. The format is as follows:
+
+```
+00001.jpg    0,0,1,0,....
+```
+
+Note 1) The images are separated by Tab[\t], 2) The annotated values are separated by commas [,]. If the format is wrong, the parsing will fail.
+
+
+### Modify The Configuration To Start Training
+
+First run the following command to download the training code (for more environmental issues, please refer to [Install_PaddleClas](https://github.com/PaddlePaddle/PaddleClas/blob/release/2.4/docs/en/installation/ install_paddleclas_en.md)):
+
+```
+git clone https://github.com/PaddlePaddle/PaddleClas
+```
+
+You need to modify the following configuration in the configuration file `PaddleClas/blob/develop/ppcls/configs/PULC/vehicle_attribute/PPLCNet_x1_0.yaml`
+
+```yaml
+DataLoader:
+  Train:
+    dataset:
+      name: MultiLabelDataset
+      image_root: "dataset/VeRi/"                     # the root path of training images
+      cls_label_path: "dataset/VeRi/train_list.txt"   # the location of the training list file
+      label_ratio: True
+      transform_ops:
+        ...
+
+  Eval:
+    dataset:
+      name: MultiLabelDataset
+      image_root: "dataset/VeRi/"                     # the root path of evaluation images
+      cls_label_path: "dataset/VeRi/val_list.txt"     # the location of the training list file
+      label_ratio: True
+      transform_ops:
+         ...
+```
+
+Note:
+
+1. here image_root path and the relative path of the image in train.txt, corresponding to the full path of the image.
+2. If you modify the number of attributes, the number of attribute types in the content configuration item should also be modified accordingly.
+
+```yaml
+# model architecture
+Arch:
+name: "PPLCNet_x1_0"
+pretrained: True
+use_ssld: True
+class_num: 19           # Number of attribute classes
+```
+
+Then run the following command to start training:
+
+```bash
+#Multi-card training
+export CUDA_VISIBLE_DEVICES=0,1,2,3
+python3 -m paddle.distributed.launch \
+    --gpus="0,1,2,3" \
+    tools/train.py \
+        -c ./ppcls/configs/PULC/vehicle_attribute/PPLCNet_x1_0.yaml
+
+#Single card training
+python3 tools/train.py \
+        -c ./ppcls/configs/PULC/vehicle_attribute/PPLCNet_x1_0.yaml
+```
+
+You can run the following commands for performance evaluation after the training is completed:
+
+```
+#Multi-card evaluation
+export CUDA_VISIBLE_DEVICES=0,1,2,3
+python3 -m paddle.distributed.launch \
+    --gpus="0,1,2,3" \
+    tools/eval.py \
+        -c ./ppcls/configs/PULC/vehicle_attribute/PPLCNet_x1_0.yaml \
+        -o Global.pretrained_model=./output/PPLCNet_x1_0/best_model
+
+#Single card evaluation
+python3 tools/eval.py \
+        -c ./ppcls/configs/PULC/vehicle_attribute/PPLCNet_x1_0.yaml \
+        -o Global.pretrained_model=./output/PPLCNet_x1_0/best_model
+```
+
+### Model Export
+
+Use the following command to export the trained model as an inference deployment model.
+
+```
+python3 tools/export_model.py \
+    -c ./ppcls/configs/PULC/vehicle_attribute/PPLCNet_x1_0.yaml \
+    -o Global.pretrained_model=output/PPLCNet_x1_0/best_model \
+    -o Global.save_inference_dir=deploy/models/PPLCNet_x1_0_vehicle_attribute_model
+```
+
+After exporting the model, if want to use it in PP-Vehicle, you need to download the [deploy infer model](https://bj.bcebos.com/v1/paddledet/models/pipeline/vehicle_attribute_model.zip) and copy `infer_cfg.yml` into the exported model folder `PPLCNet_x1_0_vehicle_attribute_model` .
+
+When you use the model, you need to modify the new model path `model_dir` entry and set `enable: True` in the configuration file of PP-Vehicle `. /deploy/pipeline/config/infer_cfg_ppvehicle.yml` .
+
+```
+VEHICLE_ATTR:
+  model_dir: [YOUR_DEPLOY_MODEL_DIR]/PPLCNet_x1_0_vehicle_attribute_infer/   #The exported model location
+  enable: True                                                              #Whether to enable the function
+```
+
+To this point,  a new attribute category recognition task is completed.
+
+
+
+## Adding or deleting attributes
+
+This is similar to the increase and decrease process of pedestrian attributes.
+
+If the attributes need to be added or deleted, you need to
+
+1) New attribute category information needs to be added or deleted when annotating the data.
+
+2) Modify the number and name of attributes used in train.txt corresponding to the training.
+
+3) Modify the training configuration, for example, the number of attributes in the ``PaddleClas/blob/develop/ppcls/configs/PULC/vehicle_attribute/PPLCNet_x1_0.yaml`` file, for details, please see the ``Modify configuration to start training`` section above.
+
+Example of adding attributes.
+
+1. Continue to add new attribute annotation values after 19 values when annotating the data.
+2. Add new attribute values to the annotated values in the train.txt file as well.
+3. The above is the annotation and training process with 19 attributes.
+
+<div width="500" align="center">
+  <img src="../../images/add_attribute.png"/>
+</div>
+
+
+The same applies to the deletion of attributes.
+
+
+## Modifications to post-processing code
+
+After modifying the attribute definition, the post-processing part of the pipeline also needs to be modified accordingly, which mainly affects the display results when the results are visualized.
+
+
+The code is at [file](https://github.com/PaddlePaddle/PaddleDetection/blob/develop/deploy/pipeline/ppvehicle/vehicle_attr.py#L108), that is, the `postprocess` function.
+
+The function implementation is described as follows:
+
+```python
+    # The name of the color/model is defined in the initialization function of the class
+    self.color_list = [
+        "yellow", "orange", "green", "gray", "red", "blue", "white",
+        "golden", "brown", "black"
+    ]
+    self.type_list = [
+        "sedan", "suv", "van", "hatchback", "mpv", "pickup", "bus", "truck",
+        "estate"
+    ]
+
+    ...
+
+    def postprocess(self, inputs, result):
+        # postprocess output of predictor
+        im_results = result['output']
+        batch_res = []
+        for res in im_results:
+            res = res.tolist()
+            attr_res = []
+            color_res_str = "Color: "
+            type_res_str = "Type: "
+            color_idx = np.argmax(res[:10])   # The first 10 items represent the color scores, and the item with the largest score is used as the color result
+            type_idx = np.argmax(res[10:])    # The last 9 items represent the model scores, and the item with the largest score is used as the model result.
+
+            # The score of color and model need to be larger than the corresponding threshold, otherwise it will be regarded as 'UnKnown'
+            if res[color_idx] >= self.color_threshold:
+                color_res_str += self.color_list[color_idx]
+            else:
+                color_res_str += "Unknown"
+            attr_res.append(color_res_str)
+
+            if res[type_idx + 10] >= self.type_threshold:
+                type_res_str += self.type_list[type_idx]
+            else:
+                type_res_str += "Unknown"
+            attr_res.append(type_res_str)
+
+            batch_res.append(attr_res)
+        result = {'output': batch_res}
+        return result
+```