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

作者您好,这是我使用自己的数据集训练出的结果,为什么迭代过程中的miou没有初始化的miou高? #17

Open
dqq813 opened this issue Oct 7, 2023 · 4 comments
Labels
question Further information is requested

Comments

@dqq813
Copy link

dqq813 commented Oct 7, 2023

dmt-voc-106-1lr__p0--l: 80.2861213684082
dmt-voc-106-1lr__p0--r: 79.29482460021973
dmt-voc-106-1lr__p1--r: 78.88628840446472
dmt-voc-106-1lr__p1--l: 79.12319898605347
dmt-voc-106-1lr__p2--r: 78.86956334114075
dmt-voc-106-1lr__p2--l: 79.26352620124817
dmt-voc-106-1lr__p3--r: 78.67986559867859
dmt-voc-106-1lr__p3--l: 78.97126078605652
dmt-voc-106-1lr__p4--r: 78.0062735080719
dmt-voc-106-1lr__p4--l: 78.25124263763428
dmt-voc-106-1lr__p5--r: 78.01116108894348
dmt-voc-106-1lr__p5--l: 77.23076343536377

@voldemortX voldemortX added the question Further information is requested label Oct 7, 2023
@voldemortX
Copy link
Owner

@dqq813 对于自己的数据集,这是有可能的。pseudo label相关方法,尤其在有标注数据很少时,泛化性不一定好。

@voldemortX
Copy link
Owner

可能要调整一下超参数

@dqq813
Copy link
Author

dqq813 commented Oct 7, 2023

用有标签数据初始化两个模型时,Loss在2.0左右稳定。然而在迭代过程中,loss只有0.01左右,且一直震荡。是不是说明两个模型预测伪标签的差异很小,模型之间没有分歧,所以就不能利用模型之间的分歧去纠正伪标签。

@voldemortX
Copy link
Owner

用有标签数据初始化两个模型时,Loss在2.0左右稳定。然而在迭代过程中,loss只有0.01左右,且一直震荡。是不是说明两个模型预测伪标签的差异很小,模型之间没有分歧,所以就不能利用模型之间的分歧去纠正伪标签。

有可能,你可以先试试普通的自训练,能不能带来提升

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

No branches or pull requests

2 participants