As described in the title, I would like to input coarse camera positions and get refined accurate camera poses from Dust3R. From my understanding of the init_from_known_poses function, it replaces the predicted pose.
Instead, I would like dust3R to use this as input and fine-tune it, is there any immediate provision for such? Or would I need to change the architecture to take in coarse pose as a conditioning and fine-tune it?
I would have a similar question about Dust3R taking in sparse pixel correspondences if provided.
As described in the title, I would like to input coarse camera positions and get refined accurate camera poses from Dust3R. From my understanding of the init_from_known_poses function, it replaces the predicted pose.
dust3r/dust3r/cloud_opt/init_im_poses.py#L24
Instead, I would like dust3R to use this as input and fine-tune it, is there any immediate provision for such? Or would I need to change the architecture to take in coarse pose as a conditioning and fine-tune it?
I would have a similar question about Dust3R taking in sparse pixel correspondences if provided.
Thanks, Suraj