Implementing Gradcam on custom FCN segmentation model

I have trained a custom FCN model using this repo, and wish to run gradcam visualizations on the detections. I have the .index .meta and the checkpoint files of my trained model. Most of the repositories are using either Keras or pytorch models, and the ones that are based only on tensorflow are for detection tasks.

I found some insight for running gradcam in segmentation tasks in this pyTorch-Gradcam repo but, I am having difficulty adapting it to my use-case.

I also explored ways to convert my tf model to Keras, but there seemed to be no way other than to re-create the model in Keras, and load the weights from the trained model.

Please guide me on how I can implement it in the mentioned FCN repo.