模型检测
关键参数
weights:训练好的模型文件
source: 检测的目标，可以是单张图片、文件夹、屏幕或者摄像头等
conf-thres: 置信度闯值，越低框越多，越高框越少
iou-thres: IOU闻值，越低框越少，越少框越多
torch.hub检测方法
安装Jupyter
命令行输入以下代码
pip install -i https://pypi.tuna.tsinghua.edu.cn/simple jupyterlab
conda install ipykernel
激活环境
python -m ipykernel install --name yolov5
新建Jupyter文件
hub_detect.ipynb
import torch#Model
model = torch.hub.load("./","yolov5s",source= "local")#Images
img = "./data/images/zidane.jpg"# Inference
results = model(img)# Results
results.show()
注意运行环境，在下yolov5
数据集构建
视频放在data02文件夹下
抽取视频帧
import cv2
import matplotlib.pyplot as plt
#%%
# 打开视频文件
video = cv2.VideoCapture("./BVN.mp4")
# 读取一帧
ret, frame = video.read()plt.imshow(frame)plt.imshow(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))video = cv2.VideoCapture("./BVN.mp4")
num = 0 # 计数器
save_step = 30 # 间隔帧
while True:ret, frame = video.read()if not ret:breaknum += 1if num % save_step == 0:cv2.imwrite("./images/" + str(num) + ".jpg", frame)
安装标准工具labelimg
pip install -i https://pypi.tuna.tsinghua.edu.cn/simple labelimg
使用labelimg
-
终端输入
labelimg -
打开需要标注的文件夹
images
- 设置保存路径
- 点击下图的标志，将保存格式改为YOLO
- 开启自动保存
- 开始标注，鼠标右键点击
Create RectBox
- 框出两个人物，并命名
-
切换下一张图片，继续标准（快捷键A下一张，D下一张，W创建RectBox）
-
标准完成后，文件夹内容如图所示
模型训练
调整文件命名
命名严格一致，不能改
文件调整
设置yaml文件
- 复制coco128.yaml文件，并命名为bvn.yaml
- 修改bvn.yaml
- 修改train.py
文件名改为bvn
workers设置为1
- 运行train.py，训练模型
中间肯能需要下载一写东西，比较慢，慢慢等（可以挂个梯子）
报错，查看下面的报错解决
报错解决
页面文件太小，无法完成操作
训练过程中，发生下图所示的报错，同时pycharm崩溃
1. 更改虚拟内存
- 进入高级系统设置，应该都会进，就不说过程了
- 设置虚拟内存大小
2. 减小占用内容大小
- 新建一个fixNvPe.py程序
# Simple script to disable ASLR and make .nv_fatb sections read-only
# Requires: pefile ( python -m pip install pefile )
# Usage: fixNvPe.py --input path/to/*.dllimport argparse
import pefile
import glob
import os
import shutildef main(args):failures = []for file in glob.glob( args.input, recursive=args.recursive ):print(f"\n---\nChecking {file}...")pe = pefile.PE(file, fast_load=True)nvbSect = [ section for section in pe.sections if section.Name.decode().startswith(".nv_fatb")]if len(nvbSect) == 1:sect = nvbSect[0]size = sect.Misc_VirtualSizeaslr = pe.OPTIONAL_HEADER.IMAGE_DLLCHARACTERISTICS_DYNAMIC_BASEwritable = 0 != ( sect.Characteristics & pefile.SECTION_CHARACTERISTICS['IMAGE_SCN_MEM_WRITE'] )print(f"Found NV FatBin! Size: {size/1024/1024:0.2f}MB ASLR: {aslr} Writable: {writable}")if (writable or aslr) and size > 0:print("- Modifying DLL")if args.backup:bakFile = f"{file}_bak"print(f"- Backing up [{file}] -> [{bakFile}]")if os.path.exists( bakFile ):print( f"- Warning: Backup file already exists ({bakFile}), not modifying file! Delete the 'bak' to allow modification")failures.append( file )continuetry:shutil.copy2( file, bakFile)except Exception as e:print( f"- Failed to create backup! [{str(e)}], not modifying file!")failures.append( file )continue# Disable ASLR for DLL, and disable writing for sectionpe.OPTIONAL_HEADER.DllCharacteristics &= ~pefile.DLL_CHARACTERISTICS['IMAGE_DLLCHARACTERISTICS_DYNAMIC_BASE']sect.Characteristics = sect.Characteristics & ~pefile.SECTION_CHARACTERISTICS['IMAGE_SCN_MEM_WRITE']try:newFile = f"{file}_mod"print( f"- Writing modified DLL to [{newFile}]")pe.write( newFile )pe.close()print( f"- Moving modified DLL to [{file}]")os.remove( file )shutil.move( newFile, file )except Exception as e:print( f"- Failed to write modified DLL! [{str(e)}]")failures.append( file )continueprint("\n\nDone!")if len(failures) > 0:print("***WARNING**** These files needed modification but failed: ")for failure in failures:print( f" - {failure}")def parseArgs():parser = argparse.ArgumentParser( description="Disable ASLR and make .nv_fatb sections read-only", formatter_class=argparse.ArgumentDefaultsHelpFormatter )parser.add_argument('--input', help="Glob to parse", default="*.dll")parser.add_argument('--backup', help="Backup modified files", default=True, required=False)parser.add_argument('--recursive', '-r', default=False, action='store_true', help="Recurse into subdirectories")return parser.parse_args()###############################
# program entry point
#
if __name__ == "__main__":args = parseArgs()main( args )
- 安装pefile
pip install -i https://pypi.tuna.tsinghua.edu.cn/simple pefile
- 终端运行fixNvPe.py文件
python fixNvPe.py --input E:\kaifa\Anaconda3\envs\yolov5\lib\site-packages\torch\lib\cudnn_adv_infer64_8.dll
intput后面的路径，就是报错那里，后面给的路径
出现下图所示表示执行完毕
RuntimeError: CUDA out of memory.
解决方式：
换小模型
AttributeError: ‘FreeTypeFont’ object has no attribute ‘getsize’
这是因为安装了新版本的 Pillow (10)，pip install tf-models-official删除了该getsize 功能，降级到 Pillow 9.5 解决了该问题
解决方式：
pip install -i https://pypi.tuna.tsinghua.edu.cn/simple Pillow==9.5
查看日志
终端输入tensorboard --logdir runs
训练效果检测
命令行输入以下命令
python detect.py --weights runs/train/exp/weights/best.pt --source data02/BVN.mp4 --view-img
