search

Parskatt / RoMa

[CVPR 2024] RoMa: Robust Dense Feature Matching; RoMa is the robust dense feature matcher capable of estimating pixel-dense warps and reliable certainties for almost any image pair.
https://parskatt.github.io/RoMa/
MIT License
556 stars 43 forks source link

what the delta_cls is used for? #11

Open lyp19 opened 10 months ago

lyp19 commented 10 months ago

and the gm_cls_loss, do not understand

Parskatt commented 10 months ago

Delta cls loss is not used. It was used in early experiments to see if regression-by-classification works for refinement.

The global matcher cls loss is the regression by classification loss for the coarse/global matcher which we do use.

lyp19 commented 10 months ago

The global matcher cls loss is different from what’s in the paper?

lyp19 commented 10 months ago

discretized into a two-dimensional grid?