You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Thank you to release the impressive work! But there is a problem that this code only consider the case that solve pnp problem with a certain query image and a certain source image.
But more generally, pnp problem might be 1 query image and many candidate images, each of which is able to index their own 3d points. In my point of view, a query image can achieve a higher accuracy when it solve pnp problem with more candidate image. So do you have any plan to release multi-view version in the future time?
The text was updated successfully, but these errors were encountered:
Hi and thanks for your interest!
The code is indeed designed for single-query relocalization, however you are free to retrieve more candidate images, merge their visible 3D points and run PnP against that cluster for increased accuracy. Note that this may increase your chances however of falling in local minima.
Thank you to release the impressive work! But there is a problem that this code only consider the case that solve pnp problem with a certain query image and a certain source image.
But more generally, pnp problem might be 1 query image and many candidate images, each of which is able to index their own 3d points. In my point of view, a query image can achieve a higher accuracy when it solve pnp problem with more candidate image. So do you have any plan to release multi-view version in the future time?
The text was updated successfully, but these errors were encountered: