Hi Lea!

Running:

self_contact_module = SelfContact(essentials_folder=essentials)
self_contact_loss = SelfContactLoss(self_contact_module, device='cuda')
self_contact_loss(vertices.to(self_contact_loss.device))

Fails with:

Traceback (most recent call last):
  File "/disk2/guytevet/.pycharm_helpers/pydev/_pydevd_bundle/pydevd_exec2.py", line 3, in Exec
    exec(exp, global_vars, local_vars)
  File "<input>", line 1, in <module>
  File "/disk2/guytevet/venvs/py3.7_vibe/lib/python3.7/site-packages/torch/nn/modules/module.py", line 1051, in _call_impl
    return forward_call(*input, **kwargs)
  File "/disk2/guytevet/selfcontact/selfcontact/losses/general_selfcontact_loss.py", line 80, in forward
    test_segments=self.test_segments
  File "/disk2/guytevet/selfcontact/selfcontact/selfcontact.py", line 214, in segment_vertices
    vertices, v2v_min, incontact, exterior, test_segments)
  File "/disk2/guytevet/selfcontact/selfcontact/selfcontact.py", line 242, in segment_hd_points
    nzv = (exp1 == exp2).any(1).reshape(-1, 3).any(1)
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!

From my understanding, this happens since the essentials are not loaded to self_contact_loss.device.

Many Thanks,

Thanks for your great work! Looking at the released 3DCP dataset - Does it contain the reported 3DCP Mocap dataset and how can I distinct 3DCP Mocap from 3DCP Scan?

Thanks:)

Hi Lea:) Applying SelfContactLoss with default parameters on batch_size=32 (i.e. [32, 10475, 3] tensor for a smplx mesh), fails for trying to allocate 112GB of GPU memory. Is there a way to handle this?

RuntimeError: CUDA out of memory. Tried to allocate 112.33 GiB (GPU 0; 10.76 GiB total capacity; 309.89 MiB already allocated; 9.30 GiB free; 346.00 MiB reserved in total by PyTorch)

Thank you for sharing so excellent research. The error happens when I run run_selfcontact_optimization.py. At first, My system is ubuntu18.04, cuda11.0, pytorch1.8.1 the detailed error is following: File "/opt/conda/envs/selfcontact/lib/python3.6/site-packages/selfcontact/utils/mesh.py", line 66, in compute_vertex_normals vertices_faces[:, 0] - vertices_faces[:, 1])) RuntimeError: index_add_cuda_ does not have a deterministic implementation, but you set 'torch.use_deterministic_algorithms(True)'. You can turn off determinism just for this operation if that's acceptable for your application. You can also file an issue at https://github.com/pytorch/pytorch/issues to help us prioritize adding deterministic support for this operation.

I set the command torch.use_deterministic_algorithms(True), but it's useless. Hoping for your answer, thanks! @paulchhuang @ahmedosman

Hi, I successfully run the run_selfcontact_optimization.py in model_type='smplx', when I change model_type from smplx to smpl, and modify some codes to be compatible with it, the mismatch still occurs. Finally, It may be a solution to utilize the smpl_to_smplx.pkl to convert to smplx by smpl model, but I don't how to use it. Could you offer some help? Thanks @paulchhuang @ahmedosman

Hi,

Thanks for proposing this meaningful work! I wonder if the running speed can be improved so that the proposed loss functions, i.e. inside, outside, normal losses, can be incorporated into a training framework?

Thanks,