Obtain object coordinates based on instance segmentation results and depth maps

edit

Hello everyone. I want to obtain the real-world coordinates of objects from the mask results obtained through instance segmentation and the corresponding depth map. Currently, I am using the mask-rcnn provided by the mmdetection toolkit. I have obtained the RGB image through ROS and successfully obtained the mask results for each object. Additionally, I have acquired the depth map corresponding to this RGB image, which is a 32-bit float representation in meters by default. How can I obtain the real-world coordinates of the objects corresponding to these masks