Accuracy too low for Mobilenet-SSD training in Caffe

Question

I have done transfer learning for object detection using the pretrained model given in the https://github.com/chuanqi305/MobileNet-SSD. These are the parameters for my scenario