loss_box: 0.0000 on custom data

[lipikak17]

Did anyone of you experience the same problem while training? the loss_box hikes up at a later part of the training sometimes but in the beginning is 0.
Background: I am training on a custom dataset with 2 classes(background + class 1). For hardware I have 2 GTX 1080 gpu's with 8GB each. Initially I had 4 custom classes and the result was a -1 map for all classes, so I changed the training by having only 1 class, as earlier it functioned quiet well for me.

Any kind of help would be appreciated!