DLT function in Eigfree code

[wahrheit-git]

Why do you normalize 'v' by the norm of third row of rotation matrix specifically and also the sign in the below line?

https://github.com/Dangzheng/Eig-Free-release/blob/74cf782d54bfa57cc66ae81bf4f2f8309f25fd70/tests.py#L1164

[kmyi]

@Dangzheng Ping!

[Dangzheng]

Hi @kk1153
The python func which you're looking at is what we convert from a Matlab script which we found on the GitHub. I forget where I find it, but I do keep the script. If you want to check I could send it to you.

The normalization part seems unnecessary to me. The sign thing seems like it wants to keep z-axis's value to be positive. Did you try to disable these line and see what happens?

Best regards,
ZhengDang

[wahrheit-git]

Can you please attach the script here? Else [email protected] . When I tried commenting this normalization, results do change. It would be great if you would be able to refer me to the source.

[zjhthu]

So will this line influence the final results? @kk1153

[wahrheit-git]

@zjhthu I could not reproduce their results. Unfortunately, the synthetic dataset used for the evaluation was also not released by the authors.

[zjhthu]

I am trying to modify the code as shown in this issue, the results seem acceptable though I only trained 10000 iterations. Still waiting for the finish of training.

[wahrheit-git]

are you evaluating for PnP (pose estimation) or homography? PnP dataset was not released, right?

[zjhthu]

I'm evaluating for the PnP problem, and generating data using the provided script.