fire-detect-yolov4 and fire-detect-dataset

• author is leilei(CSU)
• fire-detection qq群: 980489677

Some details

• darknet-yolov4-install-tutorial
• latest-fire-dataset (with xml annotations) download: BaiDuYunPan 提取码->(3q4r)
• fire-dataset (with xml annotations) download: BaiDuYunPan 提取码->(9mr7)
• fire-yolov4-weights download: BaiDuYunPan 提取码->(w3ip)
• yolov4.conv.137 -> GoogleDriver download: yolov4.conv.137

How to use dataset?

• We annotate the fire-detection-dataset as Pascal VOC format:

  --VOC2020
      --Annotations (xml_num: 2059)
      --ImageSets(Main)
      --JPEGImages (image_num: 2059)
      
      --label_name: fire
  

• If you want to convert VOC to YOLO format:

  Call darknet-yolov4's scripts voc_label.py
  

• Fire scene:

  vehicle-fire、grassland-fire、forest-fire、building-fire、Big and small fire、Day and night fire;
  

How to use this code (test)?

0. installed darknet-yolov4, and put darknet_API.py into ./darknet
1. put cfg into ./darknet
2. download fire-yolov4's weight, and put it in backup_fire folder
3. Call the darknet_API main function:

   from darknet_API import Detect
   detect = Detect(metaPath=r'./cfg/fire.data', configPath=r'./cfg/yolov4-fire.cfg',\
                   weightPath=r'./backup_fire/yolov4-fire_best.weights',\
                   namesPath=r'./cfg/fire.names')
   image = cv2.imread(r'/home/Datasets/20200714085948.jpg', -1)
   draw_img = detect.predict_image(image, save_path='./pred.jpg')
   

• Note:
  • This project should be placed in the ./darknet folder;
  • Fire generally coexists with smoke, but we only marked fire;
  • In addition, it is easy to confuse the negative sample of the fire and the light;

How to train yolov4 in darknet (train)?

0. Convert VOC format data to YOLO format data
1. Configure file information such as cfg
2. Call the darknet command:

   ./darknet detector train cfg/fire.data cfg/yolov4-fire.cfg yolov4.conv.137 -gpus 0 -map -dont_show
   

demo

• ./result: fire-detect demos

• ./xml_lab: fire-detection image annotations

Cite

• train_data contain 1-2-3-4:
• train_data1: https://blog.csdn.net/LEILEI18A/article/details/107334474
• train_data2: https://bitbucket.org/gbdi/bowfire-dataset/downloads/
• train_data3: https://github.com/OlafenwaMoses/FireNET/releases/download/v1.0/fire-dataset.zip
• train_data4: https://github.com/cair/Fire-Detection-Image-Dataset/blob/master/Fire%20images.rar
• fire-demo-dataset: http://signal.ee.bilkent.edu.tr/VisiFire/Demo/SampleClips.html

Reputation:

• This data set contains 2 parts:
• (1) Images crawled by myself, marked by myself
• (2) The data that others open source, some have annotations, some have no annotations (I re-annotate it)

other

• building-segmentation-dataset