The P_loss in detection.py

Hi, @Peterande,

Thanks for your great work, the code is very clear and reproducible, but I have some questions about the P_loss in your code,

just like the code

```python
self.smooth_loss(P, 3)
```

It seems that the p_loss do not involved in the overall loss,  and I'd like to ask what does this code do?

Looking forward for your reply, thanks!