Question about face-specific action analysis in video input

[YoungjaeDev]

Hello,

I have a question regarding the CogVLM2 model. If I provide a face bounding box for a specific person in the video (either for the initial frame or on a frame-by-frame basis), could the model perform action analysis specifically for that individual?

Also, has the model been trained on similar tasks in its supervised fine-tuning (SFT) process?

Thank you