Results of pretrained weights are different from the paper.

Question

Results of pretrained weights are different from the paper.

Opened this issue 3 years ago · 3 comments

Hi Yang,

I tried to run the pretrained weights on 4DMatch/4DLoMatch, but the results are lower than those reported in the paper. Do I need more modifications to get the results in the paper?

Answer 1 · 2022-03-07T10:19:55.000Z

Just change the line-53 in configs/test/4dmatch.yaml
to max_condition_num: 40. It should be fine.

Answer 2 · 2022-03-09T12:46:17.000Z

Yes, it is fine now. But how does max_condition_num affect the results?

Answer 3 · 2022-03-09T13:36:13.000Z

It is a threshold for the repositioning. Previously it was set to 0, which shuts down the repositioning. Thus the results was worse.