diff --git a/README.md b/README.md
index 362112c..e5313fc 100644
--- a/README.md
+++ b/README.md
@@ -113,8 +113,9 @@ PaddleRS是xxx、xxx、xxx等遥感科研院所共同基于飞桨开发的遥感
       <td>
         <b>数据格式转换</b><br>
         <ul>
-          <li>geojson to mask</li>
+          <li>coco to mask</li>
           <li>mask to shpfile</li>
+          <li>mask to geojson</li>
         </ul>
         <b>数据预处理</b><br>
         <ul>
diff --git a/docs/data/tools.md b/docs/data/tools.md
index 179188d..93aba49 100644
--- a/docs/data/tools.md
+++ b/docs/data/tools.md
@@ -2,8 +2,9 @@
 
 工具箱位于`tools`文件夹下，目前有如下工具：
 
-- `geojson2mask`：用于将geojson格式的分割标注标签转换为png格式。
+- `coco2mask`：用于将geojson格式的分割标注标签转换为png格式。
 - `mask2shp`：用于对推理得到的png提取shapefile。
+- `mask2geojson`：用于对推理得到的png提取geojson。
 - `matcher`：用于在推理前匹配两个时段的影响。
 - `spliter`：用于将大图数据进行分割以作为训练数据。
 
@@ -18,12 +19,12 @@ git clone https://github.com/PaddleCV-SIG/PaddleRS.git
 dc PaddleRS\tools
 ```
 
-### geojson2mask
+### coco2mask
 
-`geojson2mask`的主要功能是将图像以及对应json格式的分割标签转换为图像与png格式的标签，结果会分别存放在`img`和`gt`两个文件夹中。相关的数据样例可以参考[中国典型城市建筑物实例数据集](https://www.scidb.cn/detail?dataSetId=806674532768153600&dataSetType=journal)。保存结果为单通道的伪彩色图像。使用代码如下：
+`coco2mask`的主要功能是将图像以及对应json格式的分割标签转换为图像与png格式的标签，结果会分别存放在`img`和`gt`两个文件夹中。相关的数据样例可以参考[中国典型城市建筑物实例数据集](https://www.scidb.cn/detail?dataSetId=806674532768153600&dataSetType=journal)。保存结果为单通道的伪彩色图像。使用代码如下：
 
 ```shell
-python geojson2mask.py --raw_folder xxx --save_folder xxx
+python coco2mask.py --raw_folder xxx --save_folder xxx
 ```
 
 其中：
@@ -42,10 +43,27 @@ python mask2shp.py --srcimg_path xxx.tif --mask_path xxx.png [--save_path output
 其中：
 
 - `srcimg_path`：原始图像的路径，需要带有地理信息，以便为生成的shapefile提供crs等信息。
+
 - `mask_path`：推理得到的png格式的标签的路径。
+
 - `save_path`：保存shapefile的路径，默认为`output`。
+
 - `ignore_index`：忽略生成shp的索引，如背景等，默认为255。
 
+### mask2geojson
+
+  `mask2geojson`的主要功能是将推理得到的png格式的分割结果转换为geojson格式。使用代码如下：
+
+  ```shell
+  python mask2geojson.py --mask_path xxx.tif --save_path xxx.json [--epsilon 0]
+  ```
+
+  其中：
+
+  - `mask_path`：推理得到的png格式的标签的路径。
+  - `save_path`：保存geojson的路径。
+  - `epsilon`：opencv的简化参数，默认为0。
+
 ### matcher
 
 ` matcher`的主要功能是在进行变化检测的推理前，匹配两期影像的位置，并将转换后的`im2`图像保存在原地址下，命名为`im2_M.tif`。使用代码如下：
diff --git a/paddlers/custom_models/cd/models/dsifn.py b/paddlers/custom_models/cd/models/dsifn.py
index f47ddfb..cb54a4c 100644
--- a/paddlers/custom_models/cd/models/dsifn.py
+++ b/paddlers/custom_models/cd/models/dsifn.py
@@ -12,6 +12,9 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+# Refer to
+# https://github.com/GeoZcx/A-deeply-supervised-image-fusion-network-for-change-detection-in-remote-sensing-images .
+
 import paddle
 import paddle.nn as nn
 import paddle.nn.functional as F
diff --git a/paddlers/custom_models/cd/models/snunet.py b/paddlers/custom_models/cd/models/snunet.py
index 7660d60..e2f9e65 100644
--- a/paddlers/custom_models/cd/models/snunet.py
+++ b/paddlers/custom_models/cd/models/snunet.py
@@ -12,6 +12,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+# Refer to https://github.com/likyoo/Siam-NestedUNet .
+
 import paddle
 import paddle.nn as nn
 import paddle.nn.functional as F
diff --git a/paddlers/custom_models/cd/models/unet_ef.py b/paddlers/custom_models/cd/models/unet_ef.py
index cff03b5..e363148 100644
--- a/paddlers/custom_models/cd/models/unet_ef.py
+++ b/paddlers/custom_models/cd/models/unet_ef.py
@@ -12,6 +12,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+# Transferred from https://github.com/rcdaudt/fully_convolutional_change_detection/blob/master/unet.py .
+
 import paddle
 import paddle.nn as nn
 import paddle.nn.functional as F
diff --git a/paddlers/custom_models/cd/models/unet_siamconc.py b/paddlers/custom_models/cd/models/unet_siamconc.py
index 3d796a8..bcc543a 100644
--- a/paddlers/custom_models/cd/models/unet_siamconc.py
+++ b/paddlers/custom_models/cd/models/unet_siamconc.py
@@ -12,6 +12,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+# Transferred from https://github.com/rcdaudt/fully_convolutional_change_detection/blob/master/siamunet_conc.py .
+
 import paddle
 import paddle.nn as nn
 import paddle.nn.functional as F
diff --git a/paddlers/custom_models/cd/models/unet_siamdiff.py b/paddlers/custom_models/cd/models/unet_siamdiff.py
index 6ec2948..4b25e97 100644
--- a/paddlers/custom_models/cd/models/unet_siamdiff.py
+++ b/paddlers/custom_models/cd/models/unet_siamdiff.py
@@ -12,6 +12,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+# Transferred from https://github.com/rcdaudt/fully_convolutional_change_detection/blob/master/siamunet_diff.py .
+
 import paddle
 import paddle.nn as nn
 import paddle.nn.functional as F
diff --git a/requirements.txt b/requirements.txt
index 156e4be..6881d7f 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -5,7 +5,6 @@ opencv-contrib-python == 4.3.0.38
 numba == 0.53.1
 scikit-learn == 0.23.2
 scikit-image >= 0.14.0
-# numpy == 1.22.3
 pandas
 scipy
 cython
@@ -18,6 +17,7 @@ openpyxl
 easydict
 munch
 natsort
+geojson
 
 # # Self installation
 # GDAL >= 3.1.3
diff --git a/tools/geojson2mask.py b/tools/coco2mask.py
similarity index 100%
rename from tools/geojson2mask.py
rename to tools/coco2mask.py
diff --git a/tools/mask2geojson.py b/tools/mask2geojson.py
new file mode 100644
index 0000000..f409293
--- /dev/null
+++ b/tools/mask2geojson.py
@@ -0,0 +1,76 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import codecs
+import cv2
+import numpy as np
+import argparse
+import geojson
+from geojson import Polygon, Feature, FeatureCollection
+from utils import Raster, Timer
+
+
+def _gt_convert(x, y, geotf):
+    x_geo = geotf[0] + x * geotf[1] + y * geotf[2]
+    y_geo = geotf[3] + x * geotf[4] + y * geotf[5]
+    return x_geo, y_geo
+
+
+@Timer
+def convert_data(mask_path, save_path, epsilon=0):
+    raster = Raster(mask_path)
+    img = raster.getArray()
+    geo_writer = codecs.open(save_path, "w", encoding="utf-8")
+    clas = np.unique(img)
+    cv2_v = (cv2.__version__.split(".")[0] == "3")
+    feats = []
+    if not isinstance(epsilon, (int, float)):
+        epsilon = 0
+    for iclas in range(1, len(clas)):
+        tmp = np.zeros_like(img).astype("uint8")
+        tmp[img == iclas] = 1
+        # TODO: Detect internal and external contour
+        results = cv2.findContours(tmp, cv2.RETR_EXTERNAL,
+                                   cv2.CHAIN_APPROX_TC89_KCOS)
+        contours = results[1] if cv2_v else results[0]
+        # hierarchys = results[2] if cv2_v else results[1]
+        if len(contours) == 0:
+            continue
+        for contour in contours:
+            contour = cv2.approxPolyDP(contour, epsilon, True)
+            polys = []
+            for point in contour:
+                x, y = point[0]
+                xg, yg = _gt_convert(x, y, raster.geot)
+                polys.append((xg, yg))
+            polys.append(polys[0])
+            feat = Feature(
+                geometry=Polygon([polys]), properties={"class": int(iclas)})
+            feats.append(feat)
+    gjs = FeatureCollection(feats)
+    geo_writer.write(geojson.dumps(gjs))
+    geo_writer.close()
+
+
+parser = argparse.ArgumentParser(description="input parameters")
+parser.add_argument("--mask_path", type=str, required=True, \
+                    help="The path of mask tif.")
+parser.add_argument("--save_path", type=str, required=True, \
+                    help="The path to save the results, file suffix is `*.json`.")
+parser.add_argument("--epsilon", type=float, default=0, \
+                    help="The CV2 simplified parameters, `0` is the default.")
+
+if __name__ == "__main__":
+    args = parser.parse_args()
+    convert_data(args.mask_path, args.save_path, args.epsilon)
diff --git a/tutorials/train/object_detection/data/.gitignore b/tutorials/train/object_detection/data/.gitignore
new file mode 100644
index 0000000..62fb76a
--- /dev/null
+++ b/tutorials/train/object_detection/data/.gitignore
@@ -0,0 +1,3 @@
+*.zip
+*.tar.gz
+sarship/
\ No newline at end of file
diff --git a/tutorials/train/object_detection/faster_rcnn.py b/tutorials/train/object_detection/faster_rcnn.py
new file mode 100644
index 0000000..16f4aff
--- /dev/null
+++ b/tutorials/train/object_detection/faster_rcnn.py
@@ -0,0 +1,98 @@
+#!/usr/bin/env python
+
+# 目标检测模型Faster R-CNN训练示例脚本
+# 执行此脚本前，请确认已正确安装PaddleRS库
+
+import os
+
+import paddlers as pdrs
+from paddlers import transforms as T
+
+# 下载文件存放目录
+DOWNLOAD_DIR = './data/sarship/'
+# 数据集存放目录
+DATA_DIR = './data/sarship/sar_ship_1/'
+# 训练集`file_list`文件路径
+TRAIN_FILE_LIST_PATH = './data/sarship/sar_ship_1/train.txt'
+# 验证集`file_list`文件路径
+EVAL_FILE_LIST_PATH = './data/sarship/sar_ship_1/valid.txt'
+# 数据集类别信息文件路径
+LABEL_LIST_PATH = './data/sarship/sar_ship_1/labels.txt'
+# 实验目录，保存输出的模型权重和结果
+EXP_DIR = './output/faster_rcnn/'
+
+# 下载和解压SAR影像舰船检测数据集
+sarship_dataset = 'https://paddleseg.bj.bcebos.com/dataset/sar_ship_1.tar.gz'
+if not os.path.exists(DATA_DIR):
+    pdrs.utils.download_and_decompress(sarship_dataset, path=DOWNLOAD_DIR)
+
+# 定义训练和验证时使用的数据变换（数据增强、预处理等）
+# 使用Compose组合多种变换方式。Compose中包含的变换将按顺序串行执行
+# API说明：https://github.com/PaddleCV-SIG/PaddleRS/blob/develop/docs/apis/transforms.md
+train_transforms = T.Compose([
+    # 对输入影像施加随机色彩扰动
+    T.RandomDistort(),
+    # 在影像边界进行随机padding
+    T.RandomExpand(),
+    # 随机裁剪，裁块大小在一定范围内变动
+    T.RandomCrop(),
+    # 随机水平翻转
+    T.RandomHorizontalFlip(),
+    # 对batch进行随机缩放，随机选择插值方式
+    T.BatchRandomResize(
+        target_sizes=[320, 352, 384, 416, 448, 480, 512, 544, 576, 608],
+        interp='RANDOM'),
+    # 影像归一化
+    T.Normalize(
+        mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+])
+
+eval_transforms = T.Compose([
+    # 使用双三次插值将输入影像缩放到固定大小
+    T.Resize(
+        target_size=608, interp='CUBIC'),
+    # 验证阶段与训练阶段的归一化方式必须相同
+    T.Normalize(
+        mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+])
+
+# 分别构建训练和验证所用的数据集
+train_dataset = pdrs.datasets.VOCDetection(
+    data_dir=DATA_DIR,
+    file_list=TRAIN_FILE_LIST_PATH,
+    label_list=LABEL_LIST_PATH,
+    transforms=train_transforms,
+    shuffle=True)
+
+eval_dataset = pdrs.datasets.VOCDetection(
+    data_dir=DATA_DIR,
+    file_list=EVAL_FILE_LIST_PATH,
+    label_list=LABEL_LIST_PATH,
+    transforms=eval_transforms,
+    shuffle=False)
+
+# 构建Faster R-CNN模型
+# 目前已支持的模型请参考：https://github.com/PaddleCV-SIG/PaddleRS/blob/develop/docs/apis/model_zoo.md
+# 模型输入参数请参考：https://github.com/PaddleCV-SIG/PaddleRS/blob/develop/paddlers/tasks/object_detector.py
+model = pdrs.tasks.FasterRCNN(num_classes=len(train_dataset.labels))
+
+# 执行模型训练
+model.train(
+    num_epochs=10,
+    train_dataset=train_dataset,
+    train_batch_size=4,
+    eval_dataset=eval_dataset,
+    # 每多少个epoch存储一次检查点
+    save_interval_epochs=5,
+    # 每多少次迭代记录一次日志
+    log_interval_steps=4,
+    save_dir=EXP_DIR,
+    # 指定预训练权重
+    pretrain_weights='COCO',
+    # 初始学习率大小
+    learning_rate=0.005,
+    # 学习率预热（learning rate warm-up）步数与初始值
+    warmup_steps=0,
+    warmup_start_lr=0.0,
+    # 是否启用VisualDL日志功能
+    use_vdl=True)
diff --git a/tutorials/train/object_detection/faster_rcnn_sar_ship.py b/tutorials/train/object_detection/faster_rcnn_sar_ship.py
deleted file mode 100644
index 9399d3e..0000000
--- a/tutorials/train/object_detection/faster_rcnn_sar_ship.py
+++ /dev/null
@@ -1,64 +0,0 @@
-import os
-import paddlers as pdrs
-from paddlers import transforms as T
-
-# download dataset
-data_dir = 'sar_ship_1'
-if not os.path.exists(data_dir):
-    dataset_url = 'https://paddleseg.bj.bcebos.com/dataset/sar_ship_1.tar.gz'
-    pdrs.utils.download_and_decompress(dataset_url, path='./')
-
-# define transforms
-train_transforms = T.Compose([
-    T.RandomDistort(),
-    T.RandomExpand(),
-    T.RandomCrop(),
-    T.RandomHorizontalFlip(),
-    T.BatchRandomResize(
-        target_sizes=[320, 352, 384, 416, 448, 480, 512, 544, 576, 608],
-        interp='RANDOM'),
-    T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
-])
-
-eval_transforms = T.Compose([
-    T.Resize(target_size=608, interp='CUBIC'),
-    T.Normalize(
-        mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
-])
-
-# define dataset
-train_file_list = os.path.join(data_dir, 'train.txt')
-val_file_list = os.path.join(data_dir, 'valid.txt')
-label_file_list = os.path.join(data_dir, 'labels.txt')
-train_dataset = pdrs.datasets.VOCDetection(
-    data_dir=data_dir,
-    file_list=train_file_list,
-    label_list=label_file_list,
-    transforms=train_transforms,
-    shuffle=True)
-
-eval_dataset = pdrs.datasets.VOCDetection(
-    data_dir=data_dir,
-    file_list=train_file_list,
-    label_list=label_file_list,
-    transforms=eval_transforms,
-    shuffle=False)
-
-# define models
-num_classes = len(train_dataset.labels)
-model = pdrs.tasks.FasterRCNN(num_classes=num_classes)
-
-# train
-model.train(
-    num_epochs=60,
-    train_dataset=train_dataset,
-    train_batch_size=2,
-    eval_dataset=eval_dataset,
-    pretrain_weights='COCO',
-    learning_rate=0.005 / 12,
-    warmup_steps=10,
-    warmup_start_lr=0.0,
-    save_interval_epochs=5,
-    lr_decay_epochs=[20, 40],
-    save_dir='output/faster_rcnn_sar_ship',
-    use_vdl=True)
diff --git a/tutorials/train/object_detection/ppyolo.py b/tutorials/train/object_detection/ppyolo.py
new file mode 100644
index 0000000..c843d6f
--- /dev/null
+++ b/tutorials/train/object_detection/ppyolo.py
@@ -0,0 +1,99 @@
+#!/usr/bin/env python
+
+# 目标检测模型PP-YOLO训练示例脚本
+# 执行此脚本前，请确认已正确安装PaddleRS库
+
+import os
+
+import paddlers as pdrs
+from paddlers import transforms as T
+
+# 下载文件存放目录
+DOWNLOAD_DIR = './data/sarship/'
+# 数据集存放目录
+DATA_DIR = './data/sarship/sar_ship_1/'
+# 训练集`file_list`文件路径
+TRAIN_FILE_LIST_PATH = './data/sarship/sar_ship_1/train.txt'
+# 验证集`file_list`文件路径
+EVAL_FILE_LIST_PATH = './data/sarship/sar_ship_1/valid.txt'
+# 数据集类别信息文件路径
+LABEL_LIST_PATH = './data/sarship/sar_ship_1/labels.txt'
+# 实验目录，保存输出的模型权重和结果
+EXP_DIR = './output/ppyolo/'
+
+# 下载和解压SAR影像舰船检测数据集
+# 若目录已存在则不重复下载
+sarship_dataset = 'https://paddleseg.bj.bcebos.com/dataset/sar_ship_1.tar.gz'
+if not os.path.exists(DATA_DIR):
+    pdrs.utils.download_and_decompress(sarship_dataset, path=DOWNLOAD_DIR)
+
+# 定义训练和验证时使用的数据变换（数据增强、预处理等）
+# 使用Compose组合多种变换方式。Compose中包含的变换将按顺序串行执行
+# API说明：https://github.com/PaddleCV-SIG/PaddleRS/blob/develop/docs/apis/transforms.md
+train_transforms = T.Compose([
+    # 对输入影像施加随机色彩扰动
+    T.RandomDistort(),
+    # 在影像边界进行随机padding
+    T.RandomExpand(),
+    # 随机裁剪，裁块大小在一定范围内变动
+    T.RandomCrop(),
+    # 随机水平翻转
+    T.RandomHorizontalFlip(),
+    # 对batch进行随机缩放，随机选择插值方式
+    T.BatchRandomResize(
+        target_sizes=[320, 352, 384, 416, 448, 480, 512, 544, 576, 608],
+        interp='RANDOM'),
+    # 影像归一化
+    T.Normalize(
+        mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+])
+
+eval_transforms = T.Compose([
+    # 使用双三次插值将输入影像缩放到固定大小
+    T.Resize(
+        target_size=608, interp='CUBIC'),
+    # 验证阶段与训练阶段的归一化方式必须相同
+    T.Normalize(
+        mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+])
+
+# 分别构建训练和验证所用的数据集
+train_dataset = pdrs.datasets.VOCDetection(
+    data_dir=DATA_DIR,
+    file_list=TRAIN_FILE_LIST_PATH,
+    label_list=LABEL_LIST_PATH,
+    transforms=train_transforms,
+    shuffle=True)
+
+eval_dataset = pdrs.datasets.VOCDetection(
+    data_dir=DATA_DIR,
+    file_list=EVAL_FILE_LIST_PATH,
+    label_list=LABEL_LIST_PATH,
+    transforms=eval_transforms,
+    shuffle=False)
+
+# 构建PP-YOLO模型
+# 目前已支持的模型请参考：https://github.com/PaddleCV-SIG/PaddleRS/blob/develop/docs/apis/model_zoo.md
+# 模型输入参数请参考：https://github.com/PaddleCV-SIG/PaddleRS/blob/develop/paddlers/tasks/object_detector.py
+model = pdrs.tasks.PPYOLO(num_classes=len(train_dataset.labels))
+
+# 执行模型训练
+model.train(
+    num_epochs=10,
+    train_dataset=train_dataset,
+    train_batch_size=4,
+    eval_dataset=eval_dataset,
+    # 每多少个epoch存储一次检查点
+    save_interval_epochs=10,
+    # 每多少次迭代记录一次日志
+    log_interval_steps=4,
+    save_dir=EXP_DIR,
+    # 指定预训练权重
+    pretrain_weights='COCO',
+    # 初始学习率大小
+    learning_rate=0.0005,
+    # 学习率预热（learning rate warm-up）步数与初始值
+    warmup_steps=0,
+    warmup_start_lr=0.0,
+    # 是否启用VisualDL日志功能
+    use_vdl=True)
diff --git a/tutorials/train/object_detection/ppyolotiny.py b/tutorials/train/object_detection/ppyolotiny.py
new file mode 100644
index 0000000..567595b
--- /dev/null
+++ b/tutorials/train/object_detection/ppyolotiny.py
@@ -0,0 +1,99 @@
+#!/usr/bin/env python
+
+# 目标检测模型PP-YOLO Tiny训练示例脚本
+# 执行此脚本前，请确认已正确安装PaddleRS库
+
+import os
+
+import paddlers as pdrs
+from paddlers import transforms as T
+
+# 下载文件存放目录
+DOWNLOAD_DIR = './data/sarship/'
+# 数据集存放目录
+DATA_DIR = './data/sarship/sar_ship_1/'
+# 训练集`file_list`文件路径
+TRAIN_FILE_LIST_PATH = './data/sarship/sar_ship_1/train.txt'
+# 验证集`file_list`文件路径
+EVAL_FILE_LIST_PATH = './data/sarship/sar_ship_1/valid.txt'
+# 数据集类别信息文件路径
+LABEL_LIST_PATH = './data/sarship/sar_ship_1/labels.txt'
+# 实验目录，保存输出的模型权重和结果
+EXP_DIR = './output/ppyolotiny/'
+
+# 下载和解压SAR影像舰船检测数据集
+# 若目录已存在则不重复下载
+sarship_dataset = 'https://paddleseg.bj.bcebos.com/dataset/sar_ship_1.tar.gz'
+if not os.path.exists(DATA_DIR):
+    pdrs.utils.download_and_decompress(sarship_dataset, path=DOWNLOAD_DIR)
+
+# 定义训练和验证时使用的数据变换（数据增强、预处理等）
+# 使用Compose组合多种变换方式。Compose中包含的变换将按顺序串行执行
+# API说明：https://github.com/PaddleCV-SIG/PaddleRS/blob/develop/docs/apis/transforms.md
+train_transforms = T.Compose([
+    # 对输入影像施加随机色彩扰动
+    T.RandomDistort(),
+    # 在影像边界进行随机padding
+    T.RandomExpand(),
+    # 随机裁剪，裁块大小在一定范围内变动
+    T.RandomCrop(),
+    # 随机水平翻转
+    T.RandomHorizontalFlip(),
+    # 对batch进行随机缩放，随机选择插值方式
+    T.BatchRandomResize(
+        target_sizes=[320, 352, 384, 416, 448, 480, 512, 544, 576, 608],
+        interp='RANDOM'),
+    # 影像归一化
+    T.Normalize(
+        mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+])
+
+eval_transforms = T.Compose([
+    # 使用双三次插值将输入影像缩放到固定大小
+    T.Resize(
+        target_size=608, interp='CUBIC'),
+    # 验证阶段与训练阶段的归一化方式必须相同
+    T.Normalize(
+        mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+])
+
+# 分别构建训练和验证所用的数据集
+train_dataset = pdrs.datasets.VOCDetection(
+    data_dir=DATA_DIR,
+    file_list=TRAIN_FILE_LIST_PATH,
+    label_list=LABEL_LIST_PATH,
+    transforms=train_transforms,
+    shuffle=True)
+
+eval_dataset = pdrs.datasets.VOCDetection(
+    data_dir=DATA_DIR,
+    file_list=EVAL_FILE_LIST_PATH,
+    label_list=LABEL_LIST_PATH,
+    transforms=eval_transforms,
+    shuffle=False)
+
+# 构建PP-YOLO Tiny模型
+# 目前已支持的模型请参考：https://github.com/PaddleCV-SIG/PaddleRS/blob/develop/docs/apis/model_zoo.md
+# 模型输入参数请参考：https://github.com/PaddleCV-SIG/PaddleRS/blob/develop/paddlers/tasks/object_detector.py
+model = pdrs.tasks.PPYOLOTiny(num_classes=len(train_dataset.labels))
+
+# 执行模型训练
+model.train(
+    num_epochs=10,
+    train_dataset=train_dataset,
+    train_batch_size=4,
+    eval_dataset=eval_dataset,
+    # 每多少个epoch存储一次检查点
+    save_interval_epochs=5,
+    # 每多少次迭代记录一次日志
+    log_interval_steps=4,
+    save_dir=EXP_DIR,
+    # 指定预训练权重
+    pretrain_weights='COCO',
+    # 初始学习率大小
+    learning_rate=0.0001,
+    # 学习率预热（learning rate warm-up）步数与初始值
+    warmup_steps=0,
+    warmup_start_lr=0.0,
+    # 是否启用VisualDL日志功能
+    use_vdl=True)
diff --git a/tutorials/train/object_detection/ppyolov2.py b/tutorials/train/object_detection/ppyolov2.py
new file mode 100644
index 0000000..c8251bb
--- /dev/null
+++ b/tutorials/train/object_detection/ppyolov2.py
@@ -0,0 +1,99 @@
+#!/usr/bin/env python
+
+# 目标检测模型PP-YOLOv2训练示例脚本
+# 执行此脚本前，请确认已正确安装PaddleRS库
+
+import os
+
+import paddlers as pdrs
+from paddlers import transforms as T
+
+# 下载文件存放目录
+DOWNLOAD_DIR = './data/sarship/'
+# 数据集存放目录
+DATA_DIR = './data/sarship/sar_ship_1/'
+# 训练集`file_list`文件路径
+TRAIN_FILE_LIST_PATH = './data/sarship/sar_ship_1/train.txt'
+# 验证集`file_list`文件路径
+EVAL_FILE_LIST_PATH = './data/sarship/sar_ship_1/valid.txt'
+# 数据集类别信息文件路径
+LABEL_LIST_PATH = './data/sarship/sar_ship_1/labels.txt'
+# 实验目录，保存输出的模型权重和结果
+EXP_DIR = './output/ppyolov2/'
+
+# 下载和解压SAR影像舰船检测数据集
+# 若目录已存在则不重复下载
+sarship_dataset = 'https://paddleseg.bj.bcebos.com/dataset/sar_ship_1.tar.gz'
+if not os.path.exists(DATA_DIR):
+    pdrs.utils.download_and_decompress(sarship_dataset, path=DOWNLOAD_DIR)
+
+# 定义训练和验证时使用的数据变换（数据增强、预处理等）
+# 使用Compose组合多种变换方式。Compose中包含的变换将按顺序串行执行
+# API说明：https://github.com/PaddleCV-SIG/PaddleRS/blob/develop/docs/apis/transforms.md
+train_transforms = T.Compose([
+    # 对输入影像施加随机色彩扰动
+    T.RandomDistort(),
+    # 在影像边界进行随机padding
+    T.RandomExpand(),
+    # 随机裁剪，裁块大小在一定范围内变动
+    T.RandomCrop(),
+    # 随机水平翻转
+    T.RandomHorizontalFlip(),
+    # 对batch进行随机缩放，随机选择插值方式
+    T.BatchRandomResize(
+        target_sizes=[320, 352, 384, 416, 448, 480, 512, 544, 576, 608],
+        interp='RANDOM'),
+    # 影像归一化
+    T.Normalize(
+        mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+])
+
+eval_transforms = T.Compose([
+    # 使用双三次插值将输入影像缩放到固定大小
+    T.Resize(
+        target_size=608, interp='CUBIC'),
+    # 验证阶段与训练阶段的归一化方式必须相同
+    T.Normalize(
+        mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+])
+
+# 分别构建训练和验证所用的数据集
+train_dataset = pdrs.datasets.VOCDetection(
+    data_dir=DATA_DIR,
+    file_list=TRAIN_FILE_LIST_PATH,
+    label_list=LABEL_LIST_PATH,
+    transforms=train_transforms,
+    shuffle=True)
+
+eval_dataset = pdrs.datasets.VOCDetection(
+    data_dir=DATA_DIR,
+    file_list=EVAL_FILE_LIST_PATH,
+    label_list=LABEL_LIST_PATH,
+    transforms=eval_transforms,
+    shuffle=False)
+
+# 构建PP-YOLOv2模型
+# 目前已支持的模型请参考：https://github.com/PaddleCV-SIG/PaddleRS/blob/develop/docs/apis/model_zoo.md
+# 模型输入参数请参考：https://github.com/PaddleCV-SIG/PaddleRS/blob/develop/paddlers/tasks/object_detector.py
+model = pdrs.tasks.PPYOLOv2(num_classes=len(train_dataset.labels))
+
+# 执行模型训练
+model.train(
+    num_epochs=10,
+    train_dataset=train_dataset,
+    train_batch_size=4,
+    eval_dataset=eval_dataset,
+    # 每多少个epoch存储一次检查点
+    save_interval_epochs=5,
+    # 每多少次迭代记录一次日志
+    log_interval_steps=4,
+    save_dir=EXP_DIR,
+    # 指定预训练权重
+    pretrain_weights='COCO',
+    # 初始学习率大小
+    learning_rate=0.0001,
+    # 学习率预热（learning rate warm-up）步数与初始值
+    warmup_steps=0,
+    warmup_start_lr=0.0,
+    # 是否启用VisualDL日志功能
+    use_vdl=True)
diff --git a/tutorials/train/object_detection/readme.md b/tutorials/train/object_detection/readme.md
deleted file mode 100644
index 7c6a4c0..0000000
--- a/tutorials/train/object_detection/readme.md
+++ /dev/null
@@ -1,28 +0,0 @@
-The detection training demo:
-* dataset: AIR-SARShip-1.0 
-* target: ship
-* model: faster_rcnn
-
-
-Run the demo:
-
-1. Install PaddleRS
-```
-git clone https://github.com/PaddleCV-SIG/PaddleRS.git
-cd PaddleRS
-pip install -r requirements.txt
-python setup.py install
-```
-
-2. Run the demo
-```
-cd tutorials/train/detection/
-
-# run training on single GPU
-export CUDA_VISIBLE_DEVICES=0
-python faster_rcnn_sar_ship.py
-
-# run traing on multi gpu
-export CUDA_VISIBLE_DEVICES=0,1
-python -m paddle.distributed.launch faster_rcnn_sar_ship.py
-```
diff --git a/tutorials/train/object_detection/yolov3.py b/tutorials/train/object_detection/yolov3.py
new file mode 100644
index 0000000..b9d73fc
--- /dev/null
+++ b/tutorials/train/object_detection/yolov3.py
@@ -0,0 +1,98 @@
+#!/usr/bin/env python
+
+# 目标检测模型YOLOv3训练示例脚本
+# 执行此脚本前，请确认已正确安装PaddleRS库
+
+import os
+
+import paddlers as pdrs
+from paddlers import transforms as T
+
+# 下载文件存放目录
+DOWNLOAD_DIR = './data/sarship/'
+# 数据集存放目录
+DATA_DIR = './data/sarship/sar_ship_1/'
+# 训练集`file_list`文件路径
+TRAIN_FILE_LIST_PATH = './data/sarship/sar_ship_1/train.txt'
+# 验证集`file_list`文件路径
+EVAL_FILE_LIST_PATH = './data/sarship/sar_ship_1/valid.txt'
+# 数据集类别信息文件路径
+LABEL_LIST_PATH = './data/sarship/sar_ship_1/labels.txt'
+# 实验目录，保存输出的模型权重和结果
+EXP_DIR = './output/yolov3/'
+
+# 下载和解压SAR影像舰船检测数据集
+# 若目录已存在则不重复下载
+sarship_dataset = 'https://paddleseg.bj.bcebos.com/dataset/sar_ship_1.tar.gz'
+if not os.path.exists(DATA_DIR):
+    pdrs.utils.download_and_decompress(sarship_dataset, path=DOWNLOAD_DIR)
+
+# 定义训练和验证时使用的数据变换（数据增强、预处理等）
+# 使用Compose组合多种变换方式。Compose中包含的变换将按顺序串行执行
+# API说明：https://github.com/PaddleCV-SIG/PaddleRS/blob/develop/docs/apis/transforms.md
+train_transforms = T.Compose([
+    # 对输入影像施加随机色彩扰动
+    T.RandomDistort(),
+    # 在影像边界进行随机padding
+    T.RandomExpand(),
+    # 随机裁剪，裁块大小在一定范围内变动
+    T.RandomCrop(),
+    # 随机水平翻转
+    T.RandomHorizontalFlip(),
+    # 对batch进行随机缩放，随机选择插值方式
+    T.BatchRandomResize(
+        target_sizes=[320, 352, 384, 416, 448, 480, 512, 544, 576, 608],
+        interp='RANDOM'),
+    # 影像归一化
+    T.Normalize(
+        mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+])
+
+eval_transforms = T.Compose([
+    # 使用双三次插值将输入影像缩放到固定大小
+    T.Resize(
+        target_size=608, interp='CUBIC'),
+    # 验证阶段与训练阶段的归一化方式必须相同
+    T.Normalize(
+        mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+])
+
+# 分别构建训练和验证所用的数据集
+train_dataset = pdrs.datasets.VOCDetection(
+    data_dir=DATA_DIR,
+    file_list=TRAIN_FILE_LIST_PATH,
+    label_list=LABEL_LIST_PATH,
+    transforms=train_transforms,
+    shuffle=True)
+
+eval_dataset = pdrs.datasets.VOCDetection(
+    data_dir=DATA_DIR,
+    file_list=EVAL_FILE_LIST_PATH,
+    label_list=LABEL_LIST_PATH,
+    transforms=eval_transforms,
+    shuffle=False)
+
+# 构建YOLOv3模型，使用DarkNet53作为backbone
+# 目前已支持的模型请参考：https://github.com/PaddleCV-SIG/PaddleRS/blob/develop/docs/apis/model_zoo.md
+# 模型输入参数请参考：https://github.com/PaddleCV-SIG/PaddleRS/blob/develop/paddlers/tasks/object_detector.py
+model = pdrs.tasks.YOLOv3(
+    num_classes=len(train_dataset.labels), backbone='DarkNet53')
+
+# 执行模型训练
+model.train(
+    num_epochs=10,
+    train_dataset=train_dataset,
+    train_batch_size=4,
+    eval_dataset=eval_dataset,
+    # 每多少个epoch存储一次检查点
+    save_interval_epochs=5,
+    # 每多少次迭代记录一次日志
+    log_interval_steps=4,
+    save_dir=EXP_DIR,
+    # 初始学习率大小
+    learning_rate=0.0001,
+    # 学习率预热（learning rate warm-up）步数与初始值
+    warmup_steps=0,
+    warmup_start_lr=0.0,
+    # 是否启用VisualDL日志功能
+    use_vdl=True)