Hello,

I'm using the HuMMan dataset and I try to project the skeleton keypoints obtained from SMPL onto Kinect color images. However, I noticed that the projected 2D keypoints do not match with the image. It appears that the depth coordinates are flipped. Could you please let me know what might be going wrong?

I'm using the following conversion formula: Jworld = R @ Jcam + T, where J is the coordinates of 3D joints, R is the rotation matrix R from camera.json, and T is the translation vector T.

Thank you for your help.

大佬，深度数据什么时候释放呀

Can you provide the action label for each sequence?

Hello,

What is the sequences fps?

Thank you for your help.

@caizhongang hello， visualizer.py代码里面输出的smpl模型输出得到的6890个proj_vertices顶点如何得到对应的眼睛或者鼻子等的2d关键点的图像坐标？
proj_vertices = perspective_projection(vertices, K, R, T)

draw on the color image

proj_vertices = np.round(proj_vertices).astype(int).tolist()
for point in proj_vertices:
color = cv2.circle(color, tuple(point), 1, (255, 255, 255), thickness=-1)

你好，请问论文中提到的Geometry-based calibration的方式，有更详细的介绍或者参考的代码吗