Regarding the issue of losses

[gyz-star]

I found that losing images during training is very poor, manifested as continuous growth or continuous decrease. What is the reason for this? Have you ever had a similar question?