search

XuyangBai / D3Feat

[TensorFlow] Official implementation of CVPR'20 oral paper - D3Feat: Joint Learning of Dense Detection and Description of 3D Local Features https://arxiv.org/abs/2003.03164
MIT License
259 stars 38 forks source link

RANSAC #20

Closed ShengyuH closed 3 years ago

ShengyuH commented 3 years ago

hi Xuyang,

For Table 2 in your paper, may I ask which RANSAC do you use for FCGF features? Is that the same one as this: https://github.com/XuyangBai/D3Feat/blob/2213212a9bb1ecce4d1b6bf65edc8a84fa2b307f/geometric_registration/evaluate.py#L93? The result here is much better than that reported in FCGF paper, I also tried this RANSAC, but can only get 83% recall. With Mutual check, I can get 86.3% recall, which is close to that reported in your work.

Best, Shengyu

XuyangBai commented 3 years ago

Hi @ShengyuH Yes, I use the mutual check before RANSAC. Besides, the stop criteria, distance threshold will also affect the result of RANSAC. I remember that FCGF didn't release their setting for RANSAC when I submitted this paper so I use my setting to get the registration recall for different descriptors.

Best, Xuyang.

ShengyuH commented 3 years ago

Hi @ShengyuH Yes, I use the mutual check before RANSAC. Besides, the stop criteria, distance threshold will also affect the result of RANSAC. I remember that FCGF didn't release their setting for RANSAC when I submitted this paper so I use my setting to get the registration recall for different descriptors.

Best, Xuyang.

Thanks! Then this makes sense to me.