SunYM2020/UA-CMDet

evaluation problem

challow0 opened this issue · 8 comments

challow0 commented

Dear author,
Thanks for your great work. When I tried to implement your result, I found the results are very odd. Can you help me to figure out why I got so high mAP values?

image

SunYM2020 commented

Hi, this evaluation result is reasonable due to our updated data annotation, and data pre-processing process, which has led to an improvement in the accuracy of the open source code compared to the results reported in the paper.

challow0 commented

Hi, this evaluation result is reasonable due to our updated data annotation, and data pre-processing process, which has led to an improvement in the accuracy of the open source code compared to the results reported in the paper.

so this mAP value is correct?

zhangshoulong commented

这个map采用的是map50还是map50-95?

zhangshoulong commented

论文中提到map采用的是map50还是map50-95?
image
image

neuhsm commented

论文中说到训练只有12个epoch，个人感觉应该没有收敛吧

Markson-Young commented

Dear author, Thanks for your great work. When I tried to implement your result, I found the results are very odd. Can you help me to figure out why I got so high mAP values?

image

请问你是使用未更新的xml标注文件进行测试的，还是使用coco格式的json文件转换得来的xml文件进行测试的。我训练UA-CMDet没能达到你那么高的精度：
image

xiaofei624 commented

请问为什么你们的json标注中有负数，负数是什么意思

leileilei2000 commented

亲爱的作者，感谢您的出色工作。当我尝试实现您的结果时，我发现结果非常奇怪。你能帮我弄清楚为什么我得到了这么高的 mAP 值吗？
image

请问你是使用未更新的xml标注文件进行测试的，还是使用coco格式的json文件转换得来的xml文件进行测试的。我训练UA-CMDet没能达到你那么高的精度： image

我也想知道这个问题，我的map只有70左右，我是用论文中的测试集test来得到的map