Open lucasjinreal opened 2 years ago
Hi @jinfagang, do you mean how to get 3d keypoints predictions from pred_cams? If so, you can see here for reference.
@JacobLiu-S Hi, more specifically, I wanna get the 3d translation of the character. I can get the accurate rotation, but the translation I don't konw how to get it. I tried get kpts3d: https://github.com/open-mmlab/mmhuman3d/blob/b68eb070a2d7f4dcd08182b7f1fdd64a7221c9ac/mmhuman3d/models/architectures/mesh_estimator.py#L777
from here, I think it should be a relative kptsin 3d (relative to root), SO i just take the 0 index as person translation in 3d, but seems not right. what's the correct way to get it?
Hi @jinfagang, Thanks for your issues. Same as #246, would you mind updating this? 😉
@jinfagang yes, please could you share this knowledge. At least in words
I might make a PR for this, adding a kpt3d related to original image as optional output.
Hi @zetyquickly,
Sorry for the late reply. You can refer to #262. Please let me know if you have any questions.
Hi, it looks like VIBE or PARE outputs pred_cams which is cam params in very frame, other than this, is pose (rotmat) of every single fame, I currently can get a good pose animation, but how should I get real 3d position from this pred_cams?