zhanghengdev/mutualguide

Localize to Classify and Classify to Localize: Mutual Guidance in Object Detection

mutualguide's Issues

请问有画框检测的代码吗

Some questions about the paper

HELLO! I have read the paper in detail, but I still don't understand the logical relationship between IOUanchor and IOUregssion, IOUamplified. I have also read your article on Zhihu, and I have roughly understood some ideas, but it seems that the logic in the paper is not quite the same, so could you please explain it briefly, thank you very much

关于用swin在voc上不收敛的问题

您好，我按照您的方法，复现了一下您的代码，然后在VOC上完全不收敛，请问是什么问题呢，您那有训练log可以提供参考一下吗

Why a higher classification score makes a higher IoUamplified?

According to the equation $I o U_{\text {amplified }}=\left(I o U_{\text {anchor }}\right)^{\frac{\sigma-p}{\sigma}}$, IoUamplified is negatively correlated with p.

Why is there no maxpooling layer after the first conv layer in ResNet and ShuffleNet?

These two backbones in this repo are different from the official design. This will cause a huge difference in the FLOPS and performance.

MG in FCOS

where can i find the model in FCOS？

关于voc数据集

你好，请问训练集与测试集的图片以及其他信息都是放在一个文件夹下是吗

就是这里取文件，没有区分训练集测试集的文件夹