Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

检测目标不完整和检测不到的问题 #516

Open
hf62580 opened this issue Dec 26, 2022 · 4 comments
Open

检测目标不完整和检测不到的问题 #516

hf62580 opened this issue Dec 26, 2022 · 4 comments

Comments

@hf62580
Copy link

hf62580 commented Dec 26, 2022

作者您好,我基于本项目训练条码和二维码的旋转目标检测,我大概准备了108张图片,然后每天图片按5度旋转生成总数7776张的图片,其中训练集占6220张,验证集占1556张,数据增强那里只关闭mosaic,其它的没有变改,做了1000 epoch训训练,训练结果如图所示
train_batch0
train_batch1
train_batch2
train_batch3
train_batch4
train_batch5
train_batch6
train_batch7
val_batch0_labels
val_batch0_pred
val_batch1_labels
val_batch1_pred
val_batch2_labels
val_batch2_pred
results
在实际检测当中,针对长的条码,容易检测不全,短的条码容易检测不出来(已经把置信度开个了平方根,并且设置0.3的置信度)
QQ截图20221226101931
能否提供一下后续改进的思路?

@checker999
Copy link

checker999 commented Dec 26, 2022

可能需要在训练的时候注意修改下超参数文件的anchor_t 调整下 最大宽高比

@hf62580
Copy link
Author

hf62580 commented Dec 26, 2022

可能需要在训练的时候注意修改下超参数文件的anchor_t 调整下 最大宽高比

好的,anchor_t我看了默认设置是4.0,这个参数跟水平目标检测的yolov5的一样,水平目标检测的yolov5没有发现有这个问题

@hf62580
Copy link
Author

hf62580 commented Dec 29, 2022

可能需要在训练的时候注意修改下超参数文件的anchor_t 调整下 最大宽高比

你好,anchor_t 调整成10,但在训练数据处理时,autoanchor,认为默认anchor值已经可以了,不需要再调整anchor值,我试过修改代码强制kmeans新的anchor值进行训练,但效果却比不上默认,您看还有什么建议吗?

@checker999
Copy link

可以考虑尝试 在训练样本中增加更多的较长 和 较短 的条码 验证集也需要加入一些

可能需要在训练的时候注意修改下超参数文件的anchor_t 调整下 最大宽高比

你好,anchor_t 调整成10,但在训练数据处理时,autoanchor,认为默认anchor值已经可以了,不需要再调整anchor值,我试过修改代码强制kmeans新的anchor值进行训练,但效果却比不上默认,您看还有什么建议吗?

可以考虑尝试 在训练样本中增加更多的较长 和 较短 的条码 验证集也需要加入一些

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants