MobileNet-V1 SSD Ideal Loss

I’m performing transfer learning on MbNet1 SSD for object detection and the best loss I’ve had so far is 2.09.
I noticed that this number goes down when I increase the size of the dataset. But I haven’t seen it below 2.09 yet.

What is an ideal loss I’m looking for? How do I figure when is the loss good enough? What is the typical value of the loss we want to aim at?