* (3) 下载`JetPack`，请参考[NVIDIA Jetson Linux Developer Guide](https://docs.nvidia.com/jetson/l4t/index.html) 中的`Preparing a Jetson Developer Kit for Use`章节内容进行刷写系统镜像。
## 下载或编译`Paddle`预测库
本文档使用`Paddle`在`JetPack4.3`上预先编译好的预测库，请根据硬件在[安装与编译 Linux 预测库](https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0-rc/guides/05_inference_deployment/inference/build_and_install_lib_cn.html) 中选择对应版本的`Paddle`预测库。
本文档使用`Paddle`在`JetPack4.3`上预先编译好的预测库，请根据硬件在[安装与编译 Linux 预测库](https://www.paddlepaddle.org.cn/documentation/docs/zh/guides/05_inference_deployment/inference/build_and_install_lib_cn.html) 中选择对应版本的`Paddle`预测库。
若需要自己在`Jetson`平台上自定义编译`Paddle`库，请参考文档[安装与编译 Linux 预测库](https://www.paddlepaddle.org.cn/documentation/docs/zh/advanced_guide/inference_deployment/inference/build_and_install_lib_cn.html) 的`NVIDIA Jetson嵌入式硬件预测库源码编译`部分内容。
PaddlePaddle C++ 预测库针对不同的`CPU`和`CUDA`版本提供了不同的预编译版本，请根据实际情况下载: [C++预测库下载列表](https://www.paddlepaddle.org.cn/documentation/docs/zh/develop/guides/05_inference_deployment/inference/build_and_install_lib_cn.html#linux)
PaddlePaddle C++ 预测库针对不同的`CPU`和`CUDA`版本提供了不同的预编译版本，请根据实际情况下载: [C++预测库下载列表](https://www.paddlepaddle.org.cn/documentation/docs/zh/guides/05_inference_deployment/inference/build_and_install_lib_cn.html#linux)
For more installation methods such as conda, docker installation, please refer to the instructions in the [installation document](https://www.paddlepaddle.org.cn/install/quick)
Please make sure that your PaddlePaddle is installed successfully and the version is not lower than the required version. Use the following command to verify.
PaddleDetection includes support for [COCO](http://cocodataset.org) and [Pascal VOC](http://host.robots.ox.ac.uk/pascal/VOC/) by default, please follow these instructions to set up the dataset.
**Create symlinks for local datasets:**
Default dataset path in config files is `dataset/coco` and `dataset/voc`, if the
datasets are already available on disk, you can simply create symlinks to
This tutorial fine-tunes a tiny dataset by pretrained detection model for users to get a model and learn PaddleDetection quickly. The model can be trained in around 20min with good performance.
-**Note: before started, need to specifiy the GPU device as follows.**
In order to enable users to quickly produce models in a short time and master the use of PaddleDetection, this tutorial uses a pre-trained detection model to finetune small data sets. A good model can be produced in a short period of time. In actual business, it is recommended that users select a suitable model configuration file for adaptation according to their needs.
-**Set GPU**
```bash
export CUDA_VISIBLE_DEVICES=0
```
## Data Preparation
Dataset refers to [Kaggle](https://www.kaggle.com/mbkinaci/fruit-images-for-object-detection), which contains 240 images in train dataset and 60 images in test dataset. Data categories are apple, orange and banana. Download [here](https://dataset.bj.bcebos.com/PaddleDetection_demo/fruit-detection.tar) and uncompress the dataset after download, script for data preparation is located at [download_fruit.py](../../dataset/fruit/download_fruit.py). Command is as follows:
Use `yolov3_mobilenet_v1` to fine-tune the model from COCO dataset.
## Prepare Dataset
The Dataset is [Kaggle dataset](https://www.kaggle.com/andrewmvd/road-sign-detection) ，Contains 877 images, 4 data categories: crosswalk, speedlimit, stop, trafficlight.
The dataset is divided into training set(contains 701 images) and test set(contains 176 images)，[download link](https://paddlemodels.bj.bcebos.com/object_detection/roadsign_voc.tar).
Meanwhile, loss and mAP can be observed on VisualDL by set `--use_vdl` and `--vdl_log_dir`. But note Python version required >= 3.5 for VisualDL.
If you want to observe the loss change curve in real time through VisualDL, add --use_vdl=true to the training command, and set the log save path through --vdl_log_dir.
**Note: VisualDL need Python>=3.5**
Please install [VisualDL](https://github.com/PaddlePaddle/VisualDL) first
