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

potsdam_vitae_rvsa_kvidff权重推理时的精度差异 #35

Open
youngbaldy opened this issue May 2, 2024 · 4 comments
Open

potsdam_vitae_rvsa_kvidff权重推理时的精度差异 #35

youngbaldy opened this issue May 2, 2024 · 4 comments

Comments

@youngbaldy
Copy link

youngbaldy commented May 2, 2024

您好,我在使用RVSA仓库中所给出的potsdam_vitae_rvsa_kvidff.pth权重进行推理时,结果有所出入,config文件除data_root以外未作修改。我的结果如下:
image
RVSA仓库中的log日志结果为 "aAcc": 0.9115, "mIoU": 0.8307, "mAcc": 0.9005, "mFscore": 0.9061, "mPrecision": 0.9124, "mRecall": 0.9005;两者所有指标差0.3%左右,我不太确定这是否可以认为两个结果是对齐的。
我的测试集是使用mmseg官方脚本对‘2_Ortho_RGB.zip'’和‘5_Labels_all.zip'’进行划分,最后得到2016张512x512的测试集。虽然我猜测与测试集划分相关,但使用rsp_r50权重进行推理时,精度是能基本对齐的。ViTAE-Transformer/RSP#15
我确实不太理解造成这种状况的原因,期待您的回复。

@DotWang
Copy link
Collaborator

DotWang commented May 2, 2024

@youngbaldy 可能是因为kvdiff的代码在上传权重后改过一点,但是这个时间有点久了,我忘了细节了,建议用非kvdiff的权重,或者用现在的代码重新训一下kvdiff的

@youngbaldy
Copy link
Author

感谢您的回复,我使用potsdam非kvdiff权重推理2016张512x512的测试集,得到的结果为"aAcc": 0.9109, "mIoU": 0.8296, "mAcc": 0.8999, "mFscore": 0.9055, "mPrecision": 0.9117, "mRecall": 0.8999,
RVSA仓库中的log日志结果为 "aAcc": 0.9122, "mIoU": 0.8313, "mAcc": 0.9007, "mFscore": 0.9064, "mPrecision": 0.9128, "mRecall": 0.9007,
两者精度相差0.1%左右,大概可以看作是对齐?因为我不太确定,非常麻烦您,期待您的回复。

@DotWang
Copy link
Collaborator

DotWang commented May 3, 2024

@youngbaldy 差不多吧,我觉得这还有可能和数据集有关系,我们用的是之前自己做项目时手动裁的数据集,不是用mmseg裁剪的,不同人的数据集裁剪方式和参数,诸如步长之类的可能也不一样,导致了结果有差异,另外,我觉得其实统一在大图上进行滑窗预测是最公平的,按照我们自己的探索,大图滑窗预测的精度也要比一个一个预测小图再评测的高,不过目前大部分人都是裁剪后的小图上测,所以我们这次也还是在小图上测的

@youngbaldy
Copy link
Author

ook 非常感谢您的回复!

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