magicleap / Atlas

Atlas: End-to-End 3D Scene Reconstruction from Posed Images

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Train my own data

xuyanging opened this issue · comments

Hi,I want to use kitti dataset to test your model, but the output of vol_04_tsdf are all 0.9999, could your give some advice about test with other dataset?

besides, I am confused about what you said

" Note that the pretrained models are trained with Z-up metric coordinates and do not generalize to other coordinates (this means that the scale and 2 axes of the orientation ambiguity of SFM must be resolved prior to using the poses)."

How should I do to make sure my test data and pose matrix are right?

@xuyanging did you figure it " Note that the pretrained models are trained with Z-up metric coordinates and do not generalize to other coordinates (this means that the scale and 2 axes of the orientation ambiguity of SFM must be resolved prior to using the poses)."

Hi,I want to use kitti dataset to test your model, but the output of vol_04_tsdf are all 0.9999, could your give some advice about test with other dataset?

besides, I am confused about what you said

" Note that the pretrained models are trained with Z-up metric coordinates and do not generalize to other coordinates (this means that the scale and 2 axes of the orientation ambiguity of SFM must be resolved prior to using the poses)."

How should I do to make sure my test data and pose matrix are right?

你好,你成功了吗,我也想用kitti试一下

@tyraut Dear:Have you solved this problem?
你好,请问你解决问题了吗?

commented

Pardon my asking ,do u have download the pretrain model R-50.pth where the page seems closed.
can u email me it if u have QwQ.Thanks a lot!!!
wanghao020110@gmail.com