Giters

caizhongang / SMPLer-X

Official Code for "SMPLer-X: Scaling Up Expressive Human Pose and Shape Estimation"

Home Page:https://caizhongang.github.io/projects/SMPLer-X/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Hand Rotation problem

liudongdong1 opened this issue · comments

commented

Hi, thanks for your great work again.
I am utilizing the SMPL-X-H32 model to process an online video featuring extensive hand movements. When attempting to transform the model's hand rotation output into BVH motion files, I've observed that the resulting hand rotations are less than ideal. Despite adjusting the orientations, which improved some frames, inconsistencies persist in the hand orientations across others.

question 1. The degree of finger bending is too slight， and for some frame, the direction of finger bending seems wrong;
image
image
question 2. i modify the direction of the above frame, it seems ok, but other frames seem wrong;
image
image
question3. The facial parameters seem not taken effect.
`class FaceRegressor(nn.Module):
def init(self, feat_dim=768):
super(FaceRegressor, self).init()
self.expr_out = make_linear_layers([feat_dim, smpl_x.expr_code_dim], relu_final=False)
self.jaw_pose_out = make_linear_layers([feat_dim, 6], relu_final=False)

def forward(self, expr_token, jaw_pose_token):
    expr_param = self.expr_out(expr_token)  # expression parameter
    jaw_pose = self.jaw_pose_out(jaw_pose_token)  # jaw pose parameter
    return expr_param, jaw_pose`
commented

Hello, I am also studying hand Rotation problem at present, could you share the code related to your movement output? Thank you very much！

commented

model hand's initial pose is different, use relaxing hand instead.

commented

Hi, the problem seems related to flat_hand_mean, smplerx assumes 'flat_hand_mean'=False. Please check this kwarg when creating the smplx model.

commented

I want to know if this problem was solved later? What is the reason?

commented

I want to know if this problem was solved later? What is the reason?

Yes, I was able to solve the problem using flat_hand_mean=False in kwarg while creating SMPLX model.