Mask R-CNN Evaluations metrics


I am following the tutorial TorchVision 0.3 Object Detection finetuning tutorial and try to train the network with customized data.

After the first epoch I obtain evaluation results and they seem to be not okay. Please refer to the image.
Bildschirmfoto 2020-05-14 um 00.29.27

What does the -1 in evaluation mean?
And what do I do wrong?

Thank you!