why my object detection training loss is very low (<1),but the precision is still 0?

Question

I am using colab to train an object detection model, please check the colab. But i found the loss is very low, the presion is still zero, what is the wrong ?

Answer 1

Fine, i should be patient, after trainiing 10000 steps, the prediction box comes out. In my case, the precision is low maybe is because i implement many tricks to simplify the model complexity and squeeze the model file size.