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 for sharing your great work. Inspired by "Acase for using rotation invariant features in state of the art feature matchers", I am attempting to integrate Steerers with LoFTR by using Steerers to rotate the descriptors before the coarse matching stage of LoFTR. However, I have not been able to generate any matches. May I ask if there are any precautions to consider when combining this work with LoFTR?
The text was updated successfully, but these errors were encountered:
That sounds like an interesting idea. Are you optimizing a steerer for a pretrained loftr-model, or optimizing everything from scratch?
Thank you so much for your response! I plan to only modify the network matching part of Loftr while keeping the steerer output fixed. Is this approach feasible?
In principle, yes. I am not sure I entirely understand though. Do you want to steer the coarse features in loftr before inputing them to the matching transformer?
Thank you for sharing your great work. Inspired by "Acase for using rotation invariant features in state of the art feature matchers", I am attempting to integrate Steerers with LoFTR by using Steerers to rotate the descriptors before the coarse matching stage of LoFTR. However, I have not been able to generate any matches. May I ask if there are any precautions to consider when combining this work with LoFTR?
The text was updated successfully, but these errors were encountered: