yolov5
yolov5
快速入门？
一次性完成全部的环境搭建？
https://www.bilibili.com/video/BV1pP41137r1/
这个教程太简陋了，什么都说不清楚。https://www.bilibili.com/video/BV1f44y187Xg/
附带有 csdn 文章的教程，正在看。
放弃了。不敢看了。太简短，也没有说明清楚更多细节。PyTorch 怎么成功安装的。都没说清楚。https://www.bilibili.com/video/BV1S5411X7FY/
郭睿提供的教程。初步完成 pytorch 环境搭建。
官方教程
https://github.com/ultralytics/yolov5/blob/master/README.zh-CN.md
初始化 具体实践
安装依赖
在 anaconda prompt 内进入到特定的目录内，然后开始输入命令安装。
称版本不满足。这里暂时跳过，到下一步骤。
权重文件？
linux 环境下常用的命令行
conda activate pytorch
cd /home/ai_rzn/code/yolov5-master/
nohup python train.py &查看指定用户的进程
ps -u root在指定目录内，查看日志。这里默认特指 /home/ai_rzn/code/yolov5-master/ 目录
cat nohup.out常用的路径
/home/ai_rzn/code/yolov5-master/business/person-drop-litter/
/home/ai_rzn/code/yolov5-master/
yolo 训练速度慢
错误
NotImplementedError: Could not run 'torchvision::nms' with arguments from the 'CUDA' backend
解决“NotImplementedError: Could not run ‘torchvision::nms‘ with arguments from the ‘CUDA‘ backend.”
NotImplementedError: Could not run 'torchvision::nms' with arguments from the 'CUDA' backend.
本地包冲突 重装 pytorch
删除包
pip uninstall torch torchvision torchaudio重新安装，不走任何节点加速
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118AssertionError: train: No labels found in XXXXX\train.cache, can not start training
因为文件组织格式错了，文件夹名称应该是强约束的。
无法使用 device 参数实现 GPU 训练 重装 pytorch
这里先新建一个新的环境，在新的环境内安装，自己挑选的，低版本的安装命令如下：
conda install pytorch==1.12.0 torchvision==0.13.0 torchaudio==0.12.0 cudatoolkit=11.6pip install torch==1.12.0+cu116 torchvision==0.13.0+cu116 torchaudio==0.12.0 --extra-index-url https://download.pytorch.org/whl/cu116这里最终选择的是 pip 命令，结果如下：
出现错误：
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
imgviz 1.7.2 requires matplotlib, which is not installed.
imgviz 1.7.2 requires PyYAML, which is not installed.
labelme 5.2.0.post4 requires colorama, which is not installed.
labelme 5.2.0.post4 requires matplotlib, which is not installed.
labelme 5.2.0.post4 requires PyYAML, which is not installed.
labelme 5.2.0.post4 requires qtpy!=1.11.2, which is not installed.尝试手动安装，但是无法保证版本对应
pip install matplotlib PyYAML colorama qtpy!=1.11.2AssertionError: Invalid CUDA '--device 1' requested, use '--device cpu' or pass valid CUDA device(s)
linux 服务器运行出错
(pytorch) [root@localhost yolov5-master]# python train.py
Traceback (most recent call last):
File "/home/ai_rzn/code/yolov5-master/train.py", line 33, in <module>
import yaml
ModuleNotFoundError: No module named 'yaml'早停机制（Early Stopping）
TODO: 待学习
Stopping training early as no improvement observed in last 100 epochs. Best results observed at epoch 110, best model saved as best.pt. To update EarlyStopping(patience=100) pass a new patience value, i.e. `python train.py --patience 300` or use `--patience 0` to disable EarlyStopping.数据增强
对于 yolov8，这里也借用和 yolov5 相同的参数。