search

xingyizhou / StarMap

StarMap for Category-Agnostic Keypoint and Viewpoint Estimation
GNU General Public License v3.0
104 stars 19 forks source link

A question about the application of this method in human pose estimation #8

Closed zhengyuezhi closed 5 years ago

zhengyuezhi commented 5 years ago

If you want to apply the method proposed in the paper in human pose estimation, how to set the data set (mpii)，thank you

xingyizhou commented 5 years ago

Hi, Thanks for your interest in our work! If you mean applying the depth map, we have provided our implementation in https://github.com/xingyizhou/pytorch-pose-hg-3d . If you are interested in the single channel heatmap and CanViewFeature, you will need to calculate the mean 2D position of MPII dataset and follow https://github.com/xingyizhou/StarMap/blob/master/lib/datasets/Pascal3D.py#L88 to create the ground truth regression target.

YoungXIAO13 commented 5 years ago

Hi,

I have a question about the decomposition of the rotation matrix into three Euler angles. Could you tell me how did you get this equation and what is the coordinate ? Thanks

xingyizhou commented 5 years ago

Hi, I haven't decomposed the matrix into tree Euler angles. The evaluation is directly done in matrix. In the resnet angle classification baseline, we compose the three angles into matrix, you can find the details at https://github.com/xingyizhou/StarMap/blob/master/tools/EvalCls.py#L20.

YoungXIAO13 commented 5 years ago

Right, I meant where comes from this equation: np.dot(RotMat('Z', theta), np.dot(RotMat('X', - (PI / 2 - elevation)), RotMat('Z', - azimuth))) Is this rotation matrix equals to the one in the projection matrix ?

xingyizhou commented 5 years ago

Hi, The equation is from the Viewpoints and Keypoints code https://github.com/shubhtuls/ViewpointsAndKeypoints/blob/1d57050601f4a3f7842b2953be524f47d4f53b36/encoding/encodePose.m#L18 .

YoungXIAO13 commented 5 years ago

ok, I see. Thanks for your reply.