05 物体检测.md · 绪韵/nvidia_jetson_nano_practice

## 物体检测

### 使用jetson inference

在已经编译好jetson inference的基础上，手动下载识别模型SSD-Mobilenet-v2并拷贝至 /jetson_inference/data/networks 目录下，进行解压
```
cd jetson_inference/data/networks
tar -zxvf SSD-Mobilenet-v2.tar.gz
cd jetson-inference/build/aarch64/bin
python3 detectnet-camera.py
```
![](https://arloseimg.oss-cn-hangzhou.aliyuncs.com/njn-imagerec1.png)

![](https://arloseimg.oss-cn-hangzhou.aliyuncs.com/njn-imagerec2.png)

### 使用yolov3

#### yolo3编译和前向预测
```
git clone https://github.com/AlexeyAB/darknet
cd darknet
wget https://pjreddie.com/media/files/yolov3.weights
wget https://pjreddie.com/media/files/yolov3-tiny.weights
vim Makefile
```
修改Makefile中的设置
```
    GPU=1
    CUDNN=1
    OPENCV=1
```
编译
```
make -j4
```
下载预训练模型并测试
```
wget https://pjreddie.com/media/files/yolov3.weights
wget https://pjreddie.com/media/files/yolov3-tiny.weights

./darknet detector test cfg/voc.data cfg/yolov3.cfg yolov3.weights -ext_output dog.jpg
./darknet detector test cfg/coco.data cfg/yolov3-tiny.cfg yolov3-tiny.weights -ext_output dog.jpg
```

#### 训练yolo3

参考：https://docs.khadas.com/vim3/HowToTrainYolo.html

训练yolo3可使用迁移学习，下载darknet53.conv.74
```
https://pjreddie.com/media/files/darknet53.conv.74
```
- 整理数据

使用xml2txt.py从类voc标注数据中生成训练集和验证集列表

使用voc_label.py生成yolo训练使用的标注数据

- 修改cfg/voc.data

主要修改数据的路径

- 修改cfg/yolov3-voc.cfg

主要修改配置从test到train，并根据自己数据中类别数目修改cfg文件中的classes和filtersnum（(classnum+5)*3）。注意，完整的yolo3模型需要修改3处，yolo3-tiny需要修改2处。

使用darknet53.conv.74迁移学习训练yolo3模型：
```
./darknet detector train cfg/user.data cfg/yolov3-tiny-user.cfg darknet53.conv.74
```
直接训练yolo3-tiny模型
```
./darknet detector train cfg/user.data cfg/yolov3-tiny-user.cfg
```

### 常见问题

### 参考链接
- [how-to-train-an-object-detection-model-easy-for-free](https://www.dlology.com/blog/how-to-train-an-object-detection-model-easy-for-free/)
- [how-to-run-ssd-mobilenet-v2-object-detection-on-jetson-nano-at-20-fps](https://www.dlology.com/blog/how-to-run-ssd-mobilenet-v2-object-detection-on-jetson-nano-at-20-fps/)
- [HowToTrainYolo](https://docs.khadas.com/vim3/HowToTrainYolo.html)
- [Github tf_jetson_nano](https://github.com/Tony607/tf_jetson_nano)
- [yolo-v3-install-and-run-yolo-on-nvidia-jetson-nano-with-gpu](https://pysource.com/2019/08/29/yolo-v3-install-and-run-yolo-on-nvidia-jetson-nano-with-gpu/)
- [Darknet YOLOv3 on Jetson Nano（日文）](https://ai4sig.org/ja/2019/06/jetson-nano-darknet-yolov3/)
- [nvidia迁移学习](https://cloud.tencent.com/developer/article/1457146)
- [玩转Jetson Nano（五）跑通yolov3](https://blog.csdn.net/beckhans/article/details/89373939)
- [Github IntelligentEdgeHOL](https://github.com/toolboc/IntelligentEdgeHOL)
- [Github jetson_nano_detection_and_tracking](https://github.com/SteveMacenski/jetson_nano_detection_and_tracking)
- [Github opendatacam 物体检测+传输](https://github.com/opendatacam/opendatacam)
- [Github tiny_yolov2_onnx_cam](https://github.com/tsutof/tiny_yolov2_onnx_cam)
- [yolo models @baiduyun pass:9uwh ](https://pan.baidu.com/s/1YdusX1ktb9C0DRs4647fEg)

绪韵/nvidia_jetson_nano_practice

简介

发行版

贡献者

近期动态

绪韵/nvidia_jetson_nano_practice .gitee-modal { width: 500px !important; }

简介

发行版

贡献者

近期动态

搜索帮助

绪韵/nvidia_jetson_nano_practice