使用yolov8训练数据集及使用中遇到的问题

d boss

已于 2023-10-23 10:46:22 修改

阅读量1.3w

点赞数 15

分类专栏： yolov8相关文章标签： YOLO

于 2023-08-30 11:51:02 首次发布

本文链接：https://blog.csdn.net/weixin_51230935/article/details/132280058

版权

yolov8相关专栏收录该内容

5 篇文章 2 订阅

订阅专栏

1.下载yolov8文件夹

下载链接

就是这个文件夹，别怕

2.yolov8模型

下载链接，我下了yolov8s.py，放在该路径E:\nfshare\yolov8\ultralytics\weights

ps：model文件类型可以是yaml，也可以是pt

3.修改yolov8.yaml文件

E:\nfshare\yolov8\ultralytics\cfg\models\v8\yolov8.yaml，就改类别数

nc: 9  # number of classes

4.新建data文件

E:\nfshare\yolov8\ultralytics\cfg\datasets\hr.yaml，这些和yolov5一样

5.修改default.yaml文件

E:\nfshare\yolov8\ultralytics\cfg\default.yaml ，就动了以下几个参数

model: weights/yolov8s.pt  # (str, optional) path to model file, i.e. yolov8n.pt, yolov8n.yaml
data:  cfg/datasets/hr.yaml # (str, optional) path to data file, i.e. coco128.yaml
epochs: 1  # (int) number of epochs to train for
amp: False
batch: 8

运行代码 yolo cfg=cfg/default.yaml

6.训练yolov8模型（train）

1.新建.py文件训练模型

在路径下新建python脚本文件\yolov8\ultralytics\demo.py，就像运行yolov5的模型一样，运行该脚本文件。

下面这些参数要怎么设置我还没懂。

#import sys
#sys.path.append("/home/yyt/nfshare/yolov8/")
from ultralytics import YOLO

# Create a new YOLO model from scratch
#model = YOLO('/home/yyt/nfshare/yolov8/ultralytics/cfg/models/v8/yolov8.yaml')

# Load a pretrained YOLO model (recommended for training)
model = YOLO('/home/yyt/nfshare/yolov8/ultralytics/weights/yolov8s.pt')

# Train the model using the 'coco128.yaml' dataset for 3 epochs
results = model.train(data='/home/yyt/nfshare/yolov8/ultralytics/cfg/datasets/hr.yaml',amp=False,epochs=2,batch=8,val=True)

# Evaluate the model's performance on the validation set
#results = model.val(data='/home/yyt/nfshare/yolov8/ultralytics/cfg/datasets/hr.yaml',amp=False,epochs=2,batch=8)

success = model.export(format='onnx')

2.运行default.yaml文件训练模型

直接输入yolo cfg=/文件的路径/default.yaml

3.命令运行直接输：

yolo task=detect mode=train model=/yolov8/ultralytics/cfg/runs/detect/train6//weights/last.pt（模型位置，模型可以是.yaml形式或者.pt) data=/yolov8/ultralytics/cfg/datasets/hr.yaml(数据集.yaml文件位置） epochs=150 save=True resume=True （后面的都是参数，具体写什么看default.yaml里面你需要改什么）

7.测试训练后的模型（test）

1.运行default.yaml

修改default.yaml,运行代码和train一样

修改mode: val，model：/runs/detect/train6/weights/last.pt ,split: test

2.代码运行

这个方法能显示test每张图片检测的结果

yolo predict model= '/home/yyt/nfshare/yolov8/ultralytics/cfg/runs/detect/train6/weights/last.pt'（模型路径） source= '/home/yyt/nfshare/zijianshujuji/image/test'（图片文件路径）

问题

1.box_loss cls_loss dfl_loss全显示为nan，map全为0

显卡问题，batch值太大

解决：在default.yaml中，改小batch值，amp改为False

2.val的box_loss、cls_loss、dfl_loss为0，train不为0

排查问题：

1.validator.py文件

为了解决问题1 result全显示为0，我删去了validator中的如下代码，恢复看看能不能跑通val

 self.args.half = self.device.type != 'cpu'  # force FP16 val during training

失败

2.amp

amp改为ture，好的这个不能改，改了又都是nan

不是我瞎改的问题

尝试：

1.在trainer.py里面搜索half关键字，把所有有.half()变为.float()

            #'model': deepcopy(de_parallel(self.model)).half(),
            'model': deepcopy(de_parallel(self.model)).float(),
            #'ema': deepcopy(self.ema.ema).half(),
            'ema': deepcopy(self.ema.ema).float(),

无效果

2.继续修改val.py,修改batch['img'].half()改为batch['img'].float()

#batch['img'] = (batch['img'].half() if self.args.half else batch['img'].float()) / 255
 batch['img'] = (batch['img'].float()) / 255

3.继续修改validator.py

不行

4.改小batch_size

batch_size=4

不行

解决：更新环境和其他安装包，pip install -r requirement.txt

3.运行default.yaml，报错 ModuleNotFoundError: No module named 'ultralytics'

(yolov5) root@xxdell:/home/yyt/nfshare/yolov8/ultralytics# yolo cfg=/home/yyt/nfshare/yolov8/ultralytics/cfg/default.yaml
Traceback (most recent call last):
  File "/home/nephilim/environment/anaconda3/envs/yolov5/bin/yolo", line 5, in <module>
    from ultralytics.cfg import entrypoint
ModuleNotFoundError: No module named 'ultralytics'

解决方法：在/home/xx/environment/anaconda3/envs/yolov5/bin/yolo文件中添加路径

sys.path.append("/home/yyt/nfshare/yolov8/")

再次运行，解决

4.训练结果runs文件保存路径改变

原本的训练结果保存在/home/yyt/nfshare/yolov8/runs 里面，即我的共享文件夹和yolov8存在的项目文件了，现在被更改了也不知道是怎么回事，前两次训练结果就没有出现。现在的文件夹路径在虚拟机的环境路径里/home/xx/environment/anaconda3/envs/yolov5/bin/runs/detect

感觉使用的不是yolov8，而是yolov5

(base) yyt@dell:/home/xx/environment/anaconda3/envs/yolov5/bin/runs/detect$ stat train
  File: train
  Size: 4096          Blocks: 8          IO Block: 4096   directory
Device: 824h/2084d    Inode: 16883842    Links: 3
Access: (0755/drwxr-xr-x)  Uid: (    0/    root)   Gid: (    0/    root)
Access: 2023-08-21 15:02:27.587039861 +0800
Modify: 2023-08-21 14:59:15.937472731 +0800
Change: 2023-08-21 14:59:15.937472731 +0800

分析：

这几次运行的都是配置default.yaml，运行python文件，正常显示。

5.断网导致训练中断，继续训练

参考http://t.csdn.cn/UZRqy

命令行直接输入代码如下，模型改为之前跑的last.pt,epochs是总的训练次数

yolo task=detect mode=train model=/home/yyt/nfshare/yolov8/ultralytics/cfg/runs/detect/train6//weights/last.pt data=/home/yyt/nfshare/yolov8/ultralytics/cfg/datasets/hr.yaml epochs=150 save=True resume=True

6.ValueError: cannot convert float NaN to integer

训练完模型后，继续跑val测试，出现以下错误。

Validating runs/detect/train/weights/best.pt...
Ultralytics YOLOv8.0.173 🚀 Python-3.9.17 torch-2.0.1+cu117 CUDA:0 (NVIDIA GeForce GTX 1660 SUPER, 5928MiB)
Model summary (fused): 168 layers, 11129067 parameters, 0 gradients
                 Class     Images  Instances      Box(P          R      mAP50  mAP50-95):   1%|▏         | 1/78 [00:00<00:26,  2.87it/s]Exception in thread Thread-25:
Traceback (most recent call last):
  File "/home/nephilim/environment/anaconda3/envs/yolov5/lib/python3.9/threading.py", line 980, in _bootstrap_inner
    self.run()
  File "/home/nephilim/environment/anaconda3/envs/yolov5/lib/python3.9/threading.py", line 917, in run
    self._target(*self._args, **self._kwargs)
  File "/home/yyt/.local/lib/python3.9/site-packages/ultralytics/utils/plotting.py", line 446, in plot_images
    annotator.box_label(box, label, color=color)
  File "/home/yyt/.local/lib/python3.9/site-packages/ultralytics/utils/plotting.py", line 116, in box_label
    self.draw.text((box[0], box[1] - h if outside else box[1]), label, fill=txt_color, font=self.font)
  File "/home/nephilim/environment/anaconda3/envs/yolov5/lib/python3.9/site-packages/PIL/ImageDraw.py", line 556, in text
    draw_text(ink)
  File "/home/nephilim/environment/anaconda3/envs/yolov5/lib/python3.9/site-packages/PIL/ImageDraw.py", line 496, in draw_text
    coord.append(int(xy[i]))
ValueError: cannot convert float NaN to integer
                 Class     Images  Instances      Box(P          R      mAP50  mAP50-95):   3%|▎         | 2/78 [00:01<00:44,  1.70it/s]Exception in thread Thread-27:
Traceback (most recent call last):
  File "/home/nephilim/environment/anaconda3/envs/yolov5/lib/python3.9/threading.py", line 980, in _bootstrap_inner
    self.run()
  File "/home/nephilim/environment/anaconda3/envs/yolov5/lib/python3.9/threading.py", line 917, in run
    self._target(*self._args, **self._kwargs)
  File "/home/yyt/.local/lib/python3.9/site-packages/ultralytics/utils/plotting.py", line 446, in plot_images
    annotator.box_label(box, label, color=color)
  File "/home/yyt/.local/lib/python3.9/site-packages/ultralytics/utils/plotting.py", line 116, in box_label
    self.draw.text((box[0], box[1] - h if outside else box[1]), label, fill=txt_color, font=self.font)
  File "/home/nephilim/environment/anaconda3/envs/yolov5/lib/python3.9/site-packages/PIL/ImageDraw.py", line 556, in text
    draw_text(ink)
  File "/home/nephilim/environment/anaconda3/envs/yolov5/lib/python3.9/site-packages/PIL/ImageDraw.py", line 496, in draw_text
    coord.append(int(xy[i]))

暂时没解决

7.关于requirements.txt安装文件的路径

这个默认安装路径一直让我很烦，安装前需要提前转到conda环境的路径下再运行命令

cd /home/111/yyt/anaconda3/envs/yolov8/lib/python3.10/site-packages(安装包路径）
pip install -r 路径/requirements.txt

参考这两篇文章http://t.csdn.cn/9edbL，http://t.csdn.cn/oJXMF，提前确定好自己的安装位置，免得找不到安装包。

8.command 'yolo' not found ,did you mean：command 'rolo' from deb rolo

类似找不到命令的情况在yolov8里很常见

参考http://t.csdn.cn/R6aQY

运行yolov8目录下的setup.py文件

python setup.py install

9.seaborn/_oldcore.py:1119:FuturwWarning:use_inf_as_na option is be removed in a future version.Cohvert inf values to NaN before operating instead:

未来预警，意思是seaborn/_oldcore.py:1119:FuturwWarning:use_inf_as_na 选项将在未来版本中删除。在操作之前将 inf 值转换为 NaN：

只要不是报错不管他。

10.AssertionError: /home/wsjdy/yyt/nfshare/yolov8/ultralytics/runs/detect/train5/weights/last.pt training to 150 epochs is finished, nothing to resume.

训练150轮结束后，想要继续训练，增加训练轮数

修改resume=True epochs=300 models=/上次训练的结果/last.pt

结果显示报错如上。

尝试：yolov8断点恢复训练及减少训练次数和增加训练次数-CSDN博客

修改ultralytics/engine/trainer.py文件

修改self.epochs=想要训练的总次数，结果失败

尝试2：Yolov8断点续训/继续训练_q1224352995的博客-CSDN博客失败

屈服了

直接把models改为last.pt,resume=False,epochs=150,在原有的基础上加训