AP and AR equals to 0 or -1

vision
1

I am following this tutorial TorchVision Object Detection Finetuning Tutorial — PyTorch Tutorials 2.2.0+cu121 documentation
While training, the cocoevaluator returns AP and AR equals to 0 or -1 on the test set.
Does someone have an understanding of what is going on ?
Did someone stepped upon this problem ?

My LR is at 0.0000000 even when i set it up to 1e-5.

here is an image :
image

EDIT : Ok so i switched to another .ipynb notebook and the problem disappeared

2

It happened to me again, solution : increase learning rate