Grad CAM for paddle OCR

[neo133]

Hi,
I was thinking of implementing Grad Cam or any Cam based explainability method for PPOCRv4 text recognition model.. But i don't know where to start. Give me an idea as to where should i start loolking in the paddleOCR repo. Where can we find the gradient flow in the code so that I can create a hook and then plot saliency maps.