Scale Issue

Then I am trying to scale a mesh Scene, it looks I have two issues:

1. The order of the Scene Geometry is inverted
2. The scale factor is off?

Using Pytorch3D

verts, faces, aux = load_obj(FILE_PATH)
device = torch.device("cuda:0")
faces_idx = faces.verts_idx.to(device)
verts = verts.to(device)
center = verts.mean(0)
verts = verts - center
scale = max(verts.abs().max(0)[0])
verts_Pytorch3D_scaled = verts / scale

mesh = trimesh.load_mesh(FILE_PATH)
verts_Trimesh_scaled = mesh.scaled(1.0/scale)

verts_Pytorch3D_scaled != verts_Trimesh_scaled??

Why is this happening? Would be possible to preserve the scaling factor and order?

What would be the best way to scale meshes in Trimesh Scene preserving order and with the right scale?

[mikedh]

Yeah the non-uniform scaling was just released yesterday: https://github.com/mikedh/trimesh/pull/1732

There was a weird caching bug and a backwards dot product preventing this from working, added a fix and test in the followup release #1742 which I’ll merge as soon as tests pass. I think this is close to working (PR’s welcome!):

In [1]: import trimesh
m 
In [2]: m = trimesh.load('models/cycloidal.3DXML')

In [3]: m.apply_transform(trimesh.transformations.random_rotation_matrix())
Out[3]: <trimesh.Scene(len(geometry)=13)>

In [4]: m.bounds[0]
Out[4]: array([-56.02267239, -40.44721501, -36.03040504])

In [5]: s = m.scaled(1/m.extents)

In [6]: s.extents
Out[6]: array([1.01144344, 1.0091659 , 1.        ])

In [7]: s.apply_translation(-s.bounds.mean(axis=0))
Out[7]: <trimesh.Scene(len(geometry)=72)>

In [8]: s.bounds
Out[8]: 
array([[-0.50572172, -0.50458295, -0.5       ],
       [ 0.50572172,  0.50458295,  0.5       ]])

[mikedh]

Hey, not sure what Pytorch is doing, but the current trimesh behavior seems like it’s not re-ordering or inverting for a simple example:

    s = trimesh.creation.box().scene()

    scaling = 1.0 / 3.0
    c = s.scaled(scaling)

    factor = (c.geometry['geometry_0'].vertices /
              s.geometry['geometry_0'].vertices)

    assert np.allclose(factor, scaling)