yolov4 火焰检测 火焰识别 代码 数据集 开源

效果

1、训练

2、训练后给出的指标

2021年8月16日08:27:56 yolov4原版

cfg配置：
[net]
# Testing
#batch=1
#subdivisions=1
# Training
batch=64
subdivisions=32
width=512
height=512
channels=3
momentum=0.949
decay=0.0005
angle=0
saturation = 1.5
exposure = 1.5
hue=.1

learning_rate=0.001
burn_in=1000
max_batches = 4000
policy=steps
steps=3200,3600
scales=.1,.1

#cutmix=1
mosaic=1

实验效果：
 (next mAP calculation at 4455 iterations)
 Last accuracy mAP@0.50 = 93.53 %, best = 93.53 %
 4000: 1.011623, 0.935361 avg loss, 0.000010 rate, 57.365000 seconds, 256000 images, 0.650891 hours left
Resizing to initial size: 512 x 512  try to allocate additional workspace_size = 52.43 MB
 CUDA allocate done!

 calculation mAP (mean average precision)...
 Detection layer: 139 - type = 28
 Detection layer: 150 - type = 28
 Detection layer: 161 - type = 28
1580
 detections_count = 4322, unique_truth_count = 1414
class_id = 0, name = fire, ap = 92.92%           (TP = 1219, FP = 119)

 for conf_thresh = 0.25, precision = 0.91, recall = 0.86, F1-score = 0.89
 for conf_thresh = 0.25, TP = 1219, FP = 119, FN = 195, average IoU = 73.72 %

 IoU threshold = 50 %, used Area-Under-Curve for each unique Recall
 mean average precision (mAP@0.50) = 0.929237, or 92.92 %
Total Detection Time: 58 Seconds

Set -points flag:
 `-points 101` for MS COCO
 `-points 11` for PascalVOC 2007 (uncomment `difficult` in voc.data)
 `-points 0` (AUC) for ImageNet, PascalVOC 2010-2012, your custom dataset

 mean_average_precision (mAP@0.50) = 0.929237
Saving weights to weights//yolov4-fire_4000.weights
Saving weights to weights//yolov4-fire_last.weights
Saving weights to weights//yolov4-fire_final.weights
If you want to train from the beginning, then use flag in the end of training command: -clear

3、视频

yolov4 火焰检测 火焰识别 效果展示 开源 [RC汽车总动员：油罐着火了]

平台如何搭建

1、配置环境

你需要搭建好能加载yolov4权重文件的环境，具体可以参考https://github.com/AlexeyAB/darknet，里面有如何安装使用的英文教程，英文不好的看中文版，也有很多写的不错的博客重糖爱好者：YOLOv4 win10 配置 + 训练自己的数据 + 测试、YOLOv4 darknet windows10版本训练自己的数据教程、等等

2、导入权重

下载我的权重文件、.data、.cfg，导入使用即可。
链接: https://pan.baidu.com/s/16w8XON4nv1_rI6bfz0IJ4w 提取码: cm5i 复制这段内容后打开百度网盘手机App，操作更方便哦

链接: https://pan.baidu.com/s/1CHRbCQBB-Ir3tIqjHeM66A 提取码: wa53 复制这段内容后打开百度网盘手机App，操作更方便哦

3、常用命令

启动训练
darknet.exe detector train cfg/fire.data cfg\yolov4-fire.cfg  yolov4.conv.137 -map

测试视频
darknet.exe detector demo cfg/fire.data cfg\yolov4-fire.cfg backup\yolov4-fire_best60+.weights -ext_output F:\火灾代码和数据集\RC汽车总动员：油罐着火了，RC消防车紧急出动！.mp4

保存
darknet.exe detector demo cfg/coco.data cfg/yolov4.cfg yolov4.weights test.mp4 -out_filename res.avi

4、全部工程和数据集文件(可选)

https://download.csdn.net/download/bicdnknjbv/21440879

另外，我之后会公开我下载的所有数据集，同时希望可以找到一些志同道合的朋友一起标注火焰烟雾目标检测数据集，请注意我的其他帖子。