This is a real-time inference implementation of fire detection on Nvidia Jetson Nano (2GB) developer kit using NannoDet.
Notice: The JetPack 4.4 production release [L4T R32.4.3] only supports PyTorch 1.6.0 or newer, due to updates in cuDNN.
Cython
termcolor
numpy
torch
torchvision
tensorboard
pycocotools
matplotlib
pyaml
opencv-python
tqdm
Recommend torch 1.6 + vision 0.7, if your have trouble with PyTorch installation on Jetson, see Nvidia Developer Forums for detail.
My camera model is Logitech C270 with default id 0, if it's not same with yours, modify camera id by hand in advance.
$ ls /dev/video*
/dev/video0
After all the dependencies were built, connect your web cam to Jetson Nano and run:
python3 livecam.py
It may takes a while to initiate. Press Esc to quit the program.
You can train your own dataset by place them in fire/tarin and fire/val, with labelImg tool, you can easily get annotation file in voc.
fire
├── train
│ ├── ann
│ │ ├── 1.xml
│ │ └── 2.xml
│ └── img
│ ├── 1.jpg
│ └── 2.jpg
└── val
├── ann
│ └── 1.xml
└── img
└── 1.jpg
NanoDet supports dataset in both voc and coco format, you should generate your own config file before training your own dataset.
save_dir: ./fire
num_classes: 1
class_names: &class_names ['fire']
train:
name: xml_dataset
img_path: ./fire/train/img
ann_path: ./fire/train/ann
input_szie: [320,320]
val:
name: xml_dataset
img_path: ./fire/val/img
ann_path: ./fire/val/ann
input_szie: [320,320]
Due to limited resources on Jetson Nano, it's recommended to train model on PC with powerful GPU and deploy inference on Jetson Nano.
python3 train.py [your_own_config.yml]
With the end of last epoch, you will get weight file in save_dir named model_last.pth by default. In my case, it takes about 2.5 hr to train 400 images on RTX2080 with batch_size = 80 and total_epochs = 160.
https://github.com/RangiLyu/nanodet
https://courses.nvidia.com/courses/course-v1:DLI+S-RX-02+V2/about