-
Notifications
You must be signed in to change notification settings - Fork 35
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
复现结果没有达到预期 #76
Comments
emmm,你这个指标确实差的比较远。 如果是8卡训练的话,确保代码版本、config、anchor_file、data pkl都和我提供的完全一致吗? |
哈咯,请问你解决了吗,我也是用8张A100训练中会出现梯度爆炸 |
@mikasa-2024 @WangZZJJ 由于代码在处理3D 尺寸过程中使用exp, 所以预测尺寸的冪很容易爆炸,建议使用较小的学习率,并且关闭训练epoch后推理的hook机制,可以复现指标 |
请问‘关闭训练epoch后推理的hook机制’是指什么,没有太明白,意思是训练中途不进行评测,仅在全部训练完成后评测一次吗? |
是的, 前期训练不稳定也不容易收敛, 评测可以放在后面离线进行, 我记得训练过程中加上--no-validate这个参数就可以了 |
好的,我之后尝试一下,感谢! |
林博您好,我做了几组实验但是不能得到预期的结果,以下是我训练的log和结果。以下实验均是用8张A100训练。
20240722_143807.log
20240719_075541.log
20240722_032236.log
请问,您有什么建议吗?感谢!
The text was updated successfully, but these errors were encountered: