YuliangXiu / ECON

[CVPR'23, Highlight] ECON: Explicit Clothed humans Optimized via Normal integration
https://xiuyuliang.cn/econ
Other
1.11k stars 107 forks source link

How to render SMPLX front and back normal if I konw the camera parameters KRT and SMPLX parameters? #102

Open fishfishson opened 1 year ago

fishfishson commented 1 year ago

Dear author,

I want to test your normal net performance with my custom multiview data. Since I have already konwn the camera parameters and SMPLX parameters of the captured human, could you pls tell me how to render the SMPLX normal so that the results are consistent with yours?

YuliangXiu commented 1 year ago

Please check infer.py#L201 to know how ECON renders the SMPL-X normal maps from SMPL-X mesh.

fishfishson commented 1 year ago

I am sorry but it seems like there are some format mismatch for pytorch3d and my opencv cameras. Could you describe the render ing process with more details? For example the parameters of my camera is KRT and how to get the right front camera and back camera?

fishfishson commented 1 year ago

Please check infer.py#L201 to know how ECON renders the SMPL-X normal maps from SMPL-X mesh.

Should I convert perspetive camera to orthgonal camera?

YuliangXiu commented 1 year ago

Yes, all the 2D projections of ECON/ICON are orthogonal or weak perspective.

Please check TestDataset.py#L186 to see the camera details to translate/scale the estimated 3D SMPL-X body.

xianrui-luo commented 10 months ago

Yes, all the 2D projections of ECON/ICON are orthogonal or weak perspective.

Please check TestDataset.py#L186 to see the camera details to translate/scale the estimated 3D SMPL-X body.

Hi I also use perspective camera and there is still some mismatch between the rendered normal and RGB. would you explain how can I find the scale and tran in "scale, tranX, tranY = preds_dict["cam"].split(1, dim=1)" to translate/scale the estimated 3D SMPL-X body. When i read smpl model there is no such thing. Thank you.