[Python] Implementing class activation maps for models in Tensorflow Object Detection API

I need to implement CAM for a model trained on the base model ssd_mobilenet_v2_320x320_coco17_tpu-8 in the pre-trained Object Detection Zoo.

How could I accesses the layers and other fields necessary to implement Grad-CAM?

I tried converting the model to other model formats (eg. Keras), where I know how to access model layers. I did not have any success with this.

Continue reading...