Object detection training loss is not going down

I’m training a custom dataset to localize calculation part in a tax-receipt. But the training loss is revolving around 0.9. Please see the screenshot of my prediction.
I’ve tried both SSD and faster_rcnn model with Tensorflow object detection API.

You’re going to have to provide more information if you want people to help you. An example image or two and some relevant parts of your code (such as how you’re creating your Learner) would be really helpful. Otherwise, there’s not enough information for people to provide feedback

Oh, I edited the post to add pictures of my prediction. Please check. I have used tensorflow object detection api.

This forum is only for questions regarding the fastai deep learning library. If your question is for Tensorflow, I’d suggest going to Stackoverflow or one of their related sites