Ways to create bounding boxes from predicted class labels

Hi Everyone,

I am trying to learn weakly-supervised approaches to generating bounding boxes from the predicted class labels. So far, I found that we can use Class Activation Maps to first generate heatmaps and then use it to create bounding boxes. An example can be found here

Does anyone know other ways to generate bounding boxes using class label predictions?

I will appreciate any help with this.

Many Thanks and
Best Regards,