How to do transform on object mesh with obj_pose data

[dqj5182]

I am trying to render the object mesh onto the image based on 3D object template, obj_pose data, and camera intrinsic. It seems that camera intrinsic is correct as I can render hand mesh correctly on the image. However, I am having trouble aligning the object mesh onto the image. May I ask how to do it?

For now, I am extracting obj pose / obj trans / obj scale with read_rtd function from readme and conduct rotation / translation / scale with below code.

obj_verts = obj_mesh.vertices
rot_mat =  trimesh.transformations.euler_matrix(obj_rot[0], obj_rot[1], obj_rot[2])[:3, :3]
obj_verts = np.dot(obj_verts, rot_mat.T)
obj_verts = obj_verts + obj_trans
obj_verts = obj_verts * obj_scale

The result from this code is as below.

[leolyliu]

Hi dqj5182,

I cannot see the image. Could you upload it again?
The "obj_scale" should not be used in the transformation. Transforming the object model to the world coordinate system can be done using:

obj_verts = np.dot(obj_verts, rot_mat.T) + obj_trans