the pre-training loss

[jancylee]

The pre-training losses are always negative (like -199.03), is that normal?

[merlinarer]

The pre-training losses are always negative (like -199.03), is that normal?

I got the negative loss also, have you solve it?

[jancylee]

Not yet.

[yinjunbo]

Have you worked it out? What should be the correct loss?

[jxhuang0508]

Anyone reproduced the results or solved the negative loss issue?

[jianglianEin]

Anyone konw something about the negative loss? Is that right?

[impiga]

Hi, do you mean the value of loss is negative?

If this is the point, this is right for PixPro. You may check the implementation of this line https://github.com/zdaxie/PixPro/blob/main/contrast/models/PixPro.py#L204 .

[xiao7199]

Hi, If I understand correctly, this regreission loss only includes the pushing force with aligning spatial corresponding pixel. So where is the repulsing force as in Eqn2?

[ZHEGG]

why negative loss?

[shiyongde]

same question.

[douyimin]

I carefully studied the code. The loss of this implementation seems different from the paper. It does not apply constractive loss.