How to covert mmdet + mmpose to a single model

[adizhol]

Hi,

I would like to use a top-down pose model (rtmpose) which accepts bbox predictions from an object detection model (rtmdet).
During python inference (topdown_demo.py) everything works.
But I would like to deploy a single model that detects bboxes using a bbox head, and reuses the feature maps for keypoint prediction with the keypoint head.
Is this possible with mmlab/mmdeploy?
Should I make a special config by adding a a new function in rtmpose head that makes predictions from feature map inputs?

Thanks