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
622 stars 51 forks source link

image resolution to get percent correct keypoints (PCK)/ EPE #66

Closed guanfang12 closed 2 weeks ago

guanfang12 commented 3 months ago

Part 4.1 of the paper mentioned that EPE was calculated. Is it calculated using the original image size or a specific resolution? If the statistical results are not based on the original image size, can you give the PCK results under the original image size setting?

Parskatt commented 3 weeks ago

For our megadeath validation set we use a fixed resolution of 560x560 (same as the training resolution). In other benchmarks it's typically defined on original images.

guanfang12 commented 2 weeks ago

Thanks for your reply.