I wanted to implement the non local module in the mask rcnn pipeline. I've already written it as a standalone module but I'm not sure how to include it in the detectron codebase so that I can jointly train the module and the model. Could someone give me a clue as to how to include it? I simply need to train an additional 4 convolutional layers (theta, phi, g, conv) but I'm unclear as to how to do it.
I wanted to implement the non local module in the mask rcnn pipeline. I've already written it as a standalone module but I'm not sure how to include it in the detectron codebase so that I can jointly train the module and the model. Could someone give me a clue as to how to include it? I simply need to train an additional 4 convolutional layers (theta, phi, g, conv) but I'm unclear as to how to do it.
Here's the non local module:
Here's the non local module: