Getting Nan value for the loss while training a SSD-Mobilenet that was using a custom loss function

Question

Implementing the SSD-Mobilenetv1 from scratch with Pascal VOC -2007 and 2012 dataset and using GPU 3060 for training in tensorflow mainly using this code - https://github.com/ManishSoni1908/Mobilenet-ssd-keras.

the error would look like

Epoch 1/10
   1/1839 [..............................] - ETA: 709:15:59 - loss: 62.0176Batch 1: Invalid loss, terminating training
   2/1839 [..............................] - ETA: 354:31:19 - loss: inf