Extract 3D joint which in frame coordinates about hierarchicalprobabilistic3dhuman HOT 1 CLOSED

Hi,

3D SMPL joint locations in the world coordinate system can be obtained from the SMPL model output by doing

pred_joints_mode = pred_smpl_output_mode.joints  # (1, 90, 3)

similar to Line 112 in predict_poseMF_shapeGaussian_net.py.

The SMPL skeleton joints are specifically the first 24 joints out of the 90 joints which are output by the SMPL model in this repo.

To project the 3D joints from world coordinates onto the image plane, you need to define a camera in SMPL world coordinates. There is an example of this in the training script train_poseMF_shapeGaussian_net.py, line 175. This will involve rotating + translating the 3D joints according to the camera extrinsics, then perspective-projecting using the intrinsics.

from hierarchicalprobabilistic3dhuman.