Question about the moiton data optimization pipeline

[jihg88]

Thank you for your outstanding work! Regarding the use of image information to optimize local and global motions, your idea is utilizing 2D reprojection loss combined with initial motion constraints to fine-tune the osx-model so that the model can produce motions that are more consistent with image information?
I now need to reproduce this part: using image information to optimize the initial action. I don’t know how to start. Can you give me some suggestions?

[linjing7]

Hi, thanks for your interest. I think you can start with local pose optimization. We have a more detailed introduction in OSX.