Pytorch3d: How to get the positions of 2d verts after render from the 3D verts

[ForrestPi]

Hi, I am now using pytorch3d on 3D face reconstruction, but there are some small problems, I cannot get 2D landmarks from 3D vertexs
the codes follows blew:

verts, faces, aux = load_obj(trg_obj)
mesh = Meshes([verts], [faces.verts_idx], face_color)
images = renderer(mesh)

but how to get 2d verts after render from the 3D verts

Thanks

[matsuren]

Hi, if you want to get the vertices on screen, you can just use transform in the rasterizer.

verts, faces, aux = load_obj(trg_obj)
mesh = Meshes([verts], [faces.verts_idx], face_color)
mesh_on_screen = renderer.rasterizer.transform(mesh)

# Visualize vetex positions on screen
verts_on_screen = mesh_on_screen.verts_packed().detach().cpu().numpy()
plt.plot(verts_on_screen[:, 0], verts_on_screen[:, 1], 'ro')

Thanks,

[nikhilaravi]

Thanks @matsuren for replying to this! Yes you can transform the 3D vertices using the rasterizer.transform which will apply the camera transformations passed into the rasterizer.

[ForrestPi]

@matsuren thanks

[gkioxari]

Closing this issue! Feel free to reopen if you have more questions.