MLP texture does not support for DLMesh

[wangjksjtu]

Thank you for the amazing work!!

I tried to enable the mlp texture for DLM - changing this line https://github.com/NVlabs/nvdiffrec/blob/main/train.py#L632 to

mat = initial_guess_material(geometry, True, FLAGS)

However, when I run the the job, it produces the following error:

iter=    0, img_loss=0.020718, reg_loss=0.000000, lr=0.00010, time=606.2 ms, rem=10.10 m
Traceback (most recent call last):
  File "/home/wangjk/programs/nvdiffrec/train.py", line 778, in <module>
    geometry, mat = optimize_mesh(glctx, geometry, mat, lgt, dataset_train, dataset_validate, FLAGS, pass_idx=0, pass_name="mesh_pass", 
  File "/home/wangjk/programs/nvdiffrec/train.py", line 542, in optimize_mesh
    total_loss.backward()
  File "/home/wangjk/anaconda3/envs/torch-ngp/lib/python3.9/site-packages/torch/_tensor.py", line 307, in backward
    torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs)
  File "/home/wangjk/anaconda3/envs/torch-ngp/lib/python3.9/site-packages/torch/autograd/__init__.py", line 154, in backward
    Variable._execution_engine.run_backward(
RuntimeError: Trying to backward through the graph a second time (or directly access saved tensors after they have already been freed). Saved intermediate values of the graph are freed when you call .backward() or autograd.grad(). Specify retain_graph=True if you need to backward through the graph a second time or if you need to access saved tensors after calling backward.

Do you have any ideas about this? Really appreciate it ;)

[jmunkberg]

See below

[jmunkberg]

Fixed in CL 564cad1e67ad78761870fbc0ccf3f4f1f5f31f86

Thanks for reporting this @wangjksjtu !

[wangjksjtu]

Thanks a lot for the prompt reply and quick fixes! @jmunkberg Really appreciate your great help ;)