Post Process 3d Assets

[adeerAI]

Hi, thank you for the wonderful code. I want to understand that, how can I post process the following as you mentioned in Usage. And, how do I run the code?

# after much training of transformer, you can now sample novel 3d assets

faces_coordinates = transformer.generate()

# (batch, num faces, vertices (3), coordinates (3))
# now post process for the generated 3d asset

[fire]

So the mesh representation you want is

vertices = torch.randn((2, 121, 3))            # (batch, num vertices, coor (3))
faces = torch.randint(0, 121, (2, 64, 3))      # (batch, num faces, vertices (3))

Like for a 3d-simplex or tetrahedra.

vertices = torch.tensor([
    [0.0, 0.0, 0.0],  # Vertex 1
    [1.0, 0.0, 0.0],  # Vertex 2
    [0.0, 1.0, 0.0],  # Vertex 3
    [0.0, 0.0, 1.0]   # Vertex 4
]).unsqueeze(0)       # Add batch dimension

faces = torch.tensor([
    [0, 1, 2],        # Face 1
    [0, 1, 3],        # Face 2
    [0, 2, 3],        # Face 3
    [1, 2, 3]         # Face 4
]).unsqueeze(0)       # Add batch dimension

[fire]

I'll put a pr with my code for review.

[MarcusLoppe]

I think there is some confusion. The output of the transformer is triangles and those contain vertexes.
Regarding the angles,area and normal's, I've already explained in the code review but to repeat; they are automatically calculated so only the vertices and faces (indices/index to the vertices array) is needed.

Some vertex can be duplicates since two triangles might use the same vertex, so you'll need to clean it up to be most memory effective.
As per below, each triangle contains a array with 3 vertices and each vertex contains the float values for XYZ (all though not sure about the order of the x y and z):

[[-0.8047,  0.1953,  0.8516],
[-0.3359,  0.8203,  0.9609],
[ 0.3672, -0.5859,  `0.0078]]

Output of 20 triangles:

output = transformer.generate()
print(output .shape)
print(output)

torch.Size([1, 20, 3, 3])
tensor([[[[-0.8047,  0.1953,  0.8516],
          [-0.3359,  0.8203,  0.9609],
          [ 0.3672, -0.5859,  0.0078]],

         [[-0.4922, -0.7578, -0.3828],
          [-0.2422,  0.2266,  0.2578],
          [-0.2578, -0.8516, -0.3516]],
...