How can I use gradcam for Contrastive Unsupervised Representations (CURL), one of the methods for reinforcement learning?

[Kawade-Takusei]

There is a method called Contrastive Unsupervised Representations for Reinforcement Learning (CURL). I would like to know how to visualize the decision criteria of the policy network of this method with gradcam. Here is the open source of CURL. https://github.com/MishaLaskin/curl