Is it possible to use proposed refinement with the predictions from OSX model. OSX seems quite generisable model. It would be cool to have a demo where refinement is applied on osx output. Or would it require further fine tuning on predictions of osx.
Thanks for appreciating our work! ScoreHMR does not use an expressive model for the face and hands.
If you want to run refinement on top of OSX, you would need to train ScoreHMR with SMPL-X fits. Training the diffusion model with SMPLer-X predictions as pseudo ground-truth could be a good start.
Thank you for quick reply, much appreciated! What if I convert SMPL-X outputs from OSX to SMPL usiunfg revese mapping defined here, then it should work without any training right?
Is it possible to use proposed refinement with the predictions from OSX model. OSX seems quite generisable model. It would be cool to have a demo where refinement is applied on osx output. Or would it require further fine tuning on predictions of osx.
thank you for the great work! Cheers Gurkirt