variables needed for gradient computation has been modified

[Dannynis]

Hey, im trying to use this great framework to do texture optimization but im facing the next isssue,
all the steps to reproduce are in this colab notebook.

https://colab.research.google.com/drive/1K6oT4DD7NroacY3jd6k-62QI7PzLGnXM?usp=sharing

key code:

mesh.materials[0].diffuse_texture = torch.nn.Parameter(mesh.materials[0].diffuse_texture)
r = kal.render.easy_render.render_mesh(camera, mesh.cuda())
(r['render'].sum()).backward()

and the error:

File "/usr/local/lib/python3.10/dist-packages/kaolin/render/lighting/sg.py", line 340, in sg_warp_specular_term
h /= torch.sqrt(_dot(h, h))
(Triggered internally at ../torch/csrc/autograd/python_anomaly_mode.cpp:111.)
return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass

Thank you!

[OfirShechter]

Having pretty much the same issue... Any solution so far?

[Caenorst]

Hi @Dannynis , @OfirShechter ,

I'm going to push a fix soon, will be in next release

[Caenorst]

I just merged the fix, will be appearing in our next release (should be this week! :) )

[Dannynis]

works great, thank you