W (joint regressor from vertex)? about vibe HOT 5 CLOSED

M(\theta,\beta) returns a mesh in pred_vertices in

from vibe.

M(\theta,\beta) returns a mesh in pred_vertices in

VIBE/lib/models/spin.py

Lines 381 to 388 in 17b1749

	pred_output = self.smpl(
	betas=pred_shape,
	body_pose=pred_rotmat[:, 3:],
	global_orient=pred_rotmat[:, :3],
	pose2rot=True,
	)
	pred_vertices = pred_output.vertices

Given a mesh to calculate the 3d key joints it a sparse Matrix J_regressor multiplication

VIBE/lib/models/spin.py

Line 393 in 17b1749

pred_joints = torch.matmul(J_regressor_batch, pred_vertices)

As illustrated in SMPL paper:

Thanks for the reply!

from vibe.

Thanks @ikvision for the answer!

from vibe.

@wangzheallen I wrote a detailed wiki on VIBE forward pass in https://ikvision.github.io/
I hope you find it useful

from vibe.

@wangzheallen I wrote a detailed wiki on VIBE forward pass in https://ikvision.github.io/
I hope you find it useful

Thanks! I will add your nice blogs to my https://github.com/wangzheallen/awesome-human-pose-estimation

from vibe.