Ratio parameter

[arielz001]

Hi,

Im trying to run the code and Im following the steps for DiligentMV dataset:)

Im using Metashape to estimate the cameras from images and I exported the cameras to xml file from this software.
So, I run the code called metashape2neus.py, but my cameras npz file are not the same that the npz file from downloaded data, are in a different scale and I set the ratio parameter to 10.
How could I estimate the cameras correctly ? and how could I get this ratio parameter for my own data ? for example my camera is 10 cm to my object, and my object size is like 3 cm, my ratio would be 3.3 ?

Thank you so much for this great work!!

Best and thanks!

This image is the camera positions from each file, red color is the camera positions that i got and the blue is from downloaded data.

[xucao-42]

Hi,
It seems the camera parameters estimated by Metashape are not reliable (if the figure corresponds to the Diligent-MV dataset).
The well-calibrated camera parameters are stored in the 'Calib_info.mat' file of the diligent-mv dataset. The camera centers of the 20 views should have the same distance from the object and be uniformly distributed on the same elevation, since the object is rotated by a turntable and the camera is fixed.

For your information, we have updated the scene normalization method with a more intuitive hyper-parameter tuning in our new work.
https://arxiv.org/abs/2507.23162
See Section 8 and Figure 16 in the supplementary for details.
Code is coming in 1~2 weeks, but it should not be difficult to implement.

Best regards

[arielz001]

Oh thank you so much for the fast answer!! Im going to implement this! Best! :)

…

*xucao-42* left a comment (CyberAgentAILab/SuperNormal#13) <#13 (comment)> Hi, It seems the camera parameters estimated by Metashape are not reliable (if the figure corresponds to the Diligent-MV dataset). The well-calibrated camera parameters are stored in the 'Calib_info.mat' file of the diligent-mv dataset. The camera centers of the 20 views should have the same distance from the object and be uniformly distributed on the same elevation, since the object is rotated by a turntable and the camera is fixed. For your information, we have updated the scene normalization method with a more intuitive hyper-parameter tuning in our new work. https://arxiv.org/abs/2507.23162 See Section 8 and Figure 16 in the supplementary for details. Code is coming in 1~2 weeks, but it should not be difficult to implement. Best regards — Reply to this email directly, view it on GitHub <#13 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/A3UW76D5MD6L4P7ARS2G4H33MCMEBAVCNFSM6AAAAACDC6YWXKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTCNJUHEYTMMBUGM> . You are receiving this because you authored the thread.Message ID: ***@***.***>