search

abhinavdhall / deepfake

Deepfakes detection
46 stars 15 forks source link

The AUC score #1

Open zerortj opened 3 years ago

zerortj commented 3 years ago

Why my "Video wise AUC score" is 0.5 always?

ZengqiangChen commented 1 year ago

Why my "Video wise AUC score" is 0.5 always?

Have you solved it?

FranklinNotOld commented 1 year ago

Why my "Video wise AUC score" is 0.5 always?

Have you solved it?

I have come up with the same question. I don't know how to solve it.

FranklinNotOld commented 1 year ago

Why my "Video wise AUC score" is 0.5 always?

Have you solved it?

I find out the reason: in test.py, the author sets the thresholds as a fixed number, and this number is incorrect!

Txr7 commented 4 months ago

为什么我的“视频明智 AUC 分数”总是 0.5？

你解决了吗？

我找出了原因：在 test.py 中，作者将 设置为固定数字，而这个数字是不正确的！thresholds

how did you solve it? bro

Maxlee2704 commented 5 days ago

On my opinion, the reason is from def train() and def test(). You can check the output of dist = torch.dist(vid_out[batch,:].view(-1). aud_out[batch,:].view(-1), 2). The range of output is not range from [0,1] as the probability of classification. I solve it by using the cosine similarity or using normalization techniques.