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.

Sign up for GitHub

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

Jump to bottom

测试所得mAP很低。 #18

Open
s20163081161 opened this issue Jun 12, 2020 · 21 comments
Open

测试所得mAP很低。 #18

s20163081161 opened this issue Jun 12, 2020 · 21 comments

Comments

@s20163081161
Copy link

@郑博士 你好。我这边运行train+virtual数据集，当运行到40个Epoch的时候，太慢了，我把它停了，然后跑测试集，但结果rank@5才不到90%，mAP才0.35，这正常吗？？会是哪里出错了呢？？
@layumi
Copy link
Owner

layumi commented Jun 12, 2020

 60 和 75 轮会降learning rate
https://github.com/layumi/AICIty-reID-2020/blob/master/pytorch/train_2020.py#L540

应该是正常的

@s20163081161
Copy link
Author

60 和 75 轮会降learning rate
https://github.com/layumi/AICIty-reID-2020/blob/master/pytorch/train_2020.py#L540
应该是正常的

但都已经训练了30多个Epoch了，rank@5还不到90%，这也正常吗？？

@Kang9779
Copy link

@郑博士 你好。我这边运行train+virtual数据集，当运行到40个Epoch的时候，太慢了，我把它停了，然后跑测试集，但结果rank@5才不到90%，mAP才0.35，这正常吗？？会是哪里出错了呢？？

你有这个比赛的数据集吗？

@s20163081161
Copy link
Author

@郑博士 你好。我这边运行train+virtual数据集，当运行到40个Epoch的时候，太慢了，我把它停了，然后跑测试集，但结果rank@5才不到90%，mAP才0.35，这正常吗？？会是哪里出错了呢？？

你有这个比赛的数据集吗？

网上申请即可

@s20163081161
Copy link
Author

@layumi 郑博士，我用真实数据集（也就是400个目标）进行训练，把80个Epoch都跑完了，但是mAP仍是只有0.47，请问这正常吗？

@layumi
Copy link
Owner

layumi commented Jun 15, 2020 •
edited
Loading

@s20163081161
是的。差不多。因为我validation用的gallery比较大。
之后你可以尝试post processing之类的，继续提升。

@s20163081161
Copy link
Author

@s20163081161
是的。差不多。因为我validation用的gallery比较大。
之后你可以尝试post processing之类的，继续提升。

好的，多谢。

@s20163081161
Copy link
Author

@s20163081161
是的。差不多。因为我validation用的gallery比较大。
之后你可以尝试post processing之类的，继续提升。

关于您所说的“gallery比较大”，我看您这里边的gallery是36900多张图片，那么多大的数量算大呢？？有没有个衡量便准啥的？？

@s20163081161 s20163081161 reopened this Jun 15, 2020
@HanwenCao
Copy link

@郑博士 你好。我这边运行train+virtual数据集，当运行到40个Epoch的时候，太慢了，我把它停了，然后跑测试集，但结果rank@5才不到90%，mAP才0.35，这正常吗？？会是哪里出错了呢？？

您好@s20163081161，请问您跑的测试集是怎么制作的？我看readme只讲了训练集的制作。谢谢！

@s20163081161
Copy link
Author

s20163081161 commented Jun 28, 2020 via email

@HanwenCao
Copy link

您好@s20163081161，打扰了，我在网上申请并得到了所有数据集。您说的mAP的测试代码是submit.py还是test_2020.py？

@s20163081161
Copy link
Author

s20163081161 commented Jun 28, 2020 via email

@HanwenCao
Copy link

test_2020.py。你跑的如何？
…
---原始邮件--- 发件人: "Hanwen Cao"<[email protected]> 发送时间: 2020年6月28日(周日) 晚上6:12 收件人: "layumi/AICIty-reID-2020"<[email protected]>; 抄送: "Mention"<[email protected]>;"s20163081161"<[email protected]>; 主题: Re: [layumi/AICIty-reID-2020] 测试所得mAP很低。 (#18) 您好@s20163081161，打扰了，我在网上申请并得到了所有数据集。您说的mAP的测试代码是submit.py还是test_2020.py？ — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

抱歉，还没跑成功 :/

@s20163081161
Copy link
Author

s20163081161 commented Jun 28, 2020 via email

@vokhidovhusan
Copy link

vokhidovhusan commented Mar 16, 2021 •
edited
Loading

@layumi @s20163081161
I am running test_2020.py and as @layumi has mentioned earlier I am going to use val for test.
But how could I use val as a test dataset?
Should i split it to gallery and query?
Should I use val dataset as gallery or both gallery and query?

@layumi
Copy link
Owner

layumi commented Mar 16, 2021

@s20163081161 test_2020 is to verify the effectiveness of the model on validation ( no offical valid set, so I splitted ). The final test result is obtained by the submission to the official test website.

@layumi
Copy link
Owner

layumi commented Mar 16, 2021 •
edited
Loading

@martianvenusian
Test set is provided by the offical aicity organizer. If you want to verify the final result, please run submit_result.py.
But they do not provide valid set, so I splitted the set to verify the model by using test_2020.py

@vokhidovhusan
Copy link

vokhidovhusan commented Mar 16, 2021 •
edited
Loading

@layumi Thank you for your support.
I have split val set to gallery and query as following if it is right way:

val/000341/
-- 000341_c010_7.jpg
-- 000341_c016_11.jpg
-- 000341_c017_5.jpg

query/000341/
-- 000341_c010_7.jpg

gallery/000341/
-- 000341_c016_11.jpg
-- 000341_c017_5.jpg

@hongrunqv
Copy link

@郑博士 你好。我这边运行train+virtual数据集，当运行到40个Epoch的时候，太慢了，我把它停了，然后跑测试集，但结果rank@5才不到90%，mAP才0.35，这正常吗？？会是哪里出错了呢？？

你好，我想问一下在prepare_2020.py中生成的gallery这个文件夹，prepare_2020.py并没有向里面添加图片，但是在后面大量用到，这个文件夹是用来做什么的，谢谢

@wanglaotou
Copy link

@layumi Thank you for your support.
I have split val set to gallery and query as following if it is right way:

val/000341/
-- 000341_c010_7.jpg
-- 000341_c016_11.jpg
-- 000341_c017_5.jpg

query/000341/
-- 000341_c010_7.jpg

gallery/000341/
-- 000341_c016_11.jpg
-- 000341_c017_5.jpg

hello, could you please tell how to split the vaild set, which has query and gallery...

@xh7682
Copy link

xh7682 commented Sep 23, 2021

@郑博士 你好。我这边运行train+virtual数据集，当运行到40个Epoch的时候，太慢了，我把它停了，然后跑测试集，但结果rank@5才不到90%，mAP才0.35，这正常吗？？会是哪里出错了呢？？

您好@s20163081161，请问您跑的测试集是怎么制作的？我看readme只讲了训练集的制作。谢谢！
您好，请问gallery这个文件夹的数据是如何生成的呢？

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
8 participants
@layumi @wanglaotou @vokhidovhusan @s20163081161 @Kang9779 @HanwenCao @xh7682 @hongrunqv