search

Parskatt / DKM

[CVPR 2023] DKM: Dense Kernelized Feature Matching for Geometry Estimation
https://parskatt.github.io/DKM/
Other
378 stars 28 forks source link

About testing results #40

Open sugar-fly opened 1 year ago

sugar-fly commented 1 year ago

Hi Johan,

When I train about 160k steps, I got the checkpoint for testing, and then obtained different testing results. For example, 1st testing result: 58.345 (auc5), 2nd testing result: 59.449, 3th testing result: 59.853. I think this is caused by the randomness of sampling. Will this randomness be alleviated as the number of iterations increases?

Thank you so much for your help!

Parskatt commented 1 year ago

Hi, the randomness comes both from the network and the ransac. I think the results are usually more even at the end of the training due to reduced learning rate.

I think your checkpoint is before the decay right?

sugar-fly commented 1 year ago

Right. I think maybe my network has not converged very well.