The following error was reported during my running demo.py: RuntimeError: Legacy autograd function with non-static forward method is deprecated，

[smallsun1122]

[q838458640]

I had the same problem, probably because the version of the Torch was too high, but I couldn't fix it.I wonder if you have solved this problem?

[smallsun1122]

同学，你好，时间太久了，我有点忘记了，我在网上找了一些解决办法，看能不能帮到你，这些解决办法我看着有点熟悉 ，可能是之前用过的，希望可以帮到你。

------------------ 原始邮件 ------------------ 发件人: "Tencent/FaceDetection-DSFD" @.>; 发送时间: 2021年4月19日(星期一) 晚上8:27 @.>; 抄送: "small @.@.>; 主题: Re: [Tencent/FaceDetection-DSFD] The following error was reported during my running demo.py: RuntimeError: Legacy autograd function with non-static forward method is deprecated， (#72)

I had the same problem, probably because the version of the Torch was too high, but I couldn't fix it.I wonder if you have solved this problem?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

[q838458640]

感谢您

[WeiCUI6]

I also had the same issue, can I ask how do you fix the problems? I will really appreciate it!

[berkanttubi]

Any answer??

[berkanttubi]

add @staticmethod to the beginning of the function that error has been occured.

[funykatebird]

I had the same problem, probably because the version of the Torch was too high, but I couldn't fix it.I wonder if you have solved this problem? 我也遇到了同样的问题，请问解决了吗？

[wengshihao]

you can easily solve this problem by: do changes in face_ssd.py:

output = self.detect(
                  face_loc.view(face_loc.size(0), -1, 4),         # loc preds
                  self.softmax(face_conf.view(face_conf.size(0), -1, self.num_classes)), # conf preds
                  self.priors.type(type(x.data))                  # default boxes
                )

to the below code:

output = self.detect.forward(
                  face_loc.view(face_loc.size(0), -1, 4),         # loc preds
                  self.softmax(face_conf.view(face_conf.size(0), -1, self.num_classes)), # conf preds
                  self.priors.type(type(x.data))                  # default boxes
                )