Krasjet-Yu / YOLO-FaceV2

YOLO-FaceV2: A Scale and Occlusion Aware Face Detector
163 stars 24 forks source link

作者您好,请问NWD损失常量参数为什么设置为1? #38

Open wenjunx opened 12 months ago

wenjunx commented 12 months ago

nwd = torch.exp(-torch.pow(Wasserstein(pbox.T, tbox[i], x1y1x2y2=False), 1 / 2) / self.C) self.C我看这个参数在超参数文件中设置为1,我看NWD论文不是说根据数据集设置吗?

Skyninth commented 11 months ago

同问,想知道这个常数是怎么根据数据集算出来的,能提交一下代码吗