cuiziteng / ICCV_MAET

🌕 [ICCV 2021] Multitask AET with Orthogonal Tangent Regularity for Dark Object Detection. A self-supervised learning way for low-light image object detection.
Apache License 2.0
142 stars 14 forks source link

有关实验结果的确认 #26

Closed Xiaoyu30star closed 5 months ago

Xiaoyu30star commented 5 months ago

作者您好!我最近学习了您MAET这篇工作的论文与代码,在实验过程中得到了一些不同的结果:MAET方法得到的结果的确有涨点;但是只在COCO数据集上进行预训练的YOLOv3在ExDark数据集上经过微调后也很容易就得到了不错的结果,也就是YOLO(N)得到了0.744的mAP,相比论文中报告的0.627似乎有一些差异,我仔细check了一下代码,在训练策略、数据处理等各部分是与源代码一致的,只是更换了加载的预训练权重文件。而且基于Kind、ZeroDCE这些方法的最后表现几乎与YOLO(N)没有区别,甚至反而有下降,所以有些不知所措了。

Xiaoyu30star commented 5 months ago

不过我是在最新的MMDetection版本上跑的,不知道是不是其相比较之前版本在训练以及一些其他方面做了改进,从而导致了结果的提高,因为我也是刚刚接触这些东西,所以想请教学习一下

cuiziteng commented 5 months ago

I'm sorry, the ICCV 2021 paper's results are changed, the renewed results I provide in the README.md

Please check ~

image
Xiaoyu30star commented 5 months ago

非常感谢,是我没有注意到README文件更新后的baseline结果,谢谢你!