mjvanderboon / headnerf Goto Github PK

Hello,

I'm trying to run the training script (train.py) on one of the datasets (specifically on comprehensive_cars), and I'm having some problems.

Just for the record, I'm doing that as a first step, since my ultimate goal is to fine-tune a pre-trained model on my own custom dataset.

After following all the installation steps and dataset downloads shown in the README.md file, I run $ python train.py configs/256res/cars_256.yaml and mainly two error/warning messages are outputted multiple times:

Cannot re-initialize CUDA in forked subprocess. To use CUDA with multiprocessing, you must use the 'spawn' start method

Warning: Error occurred when loading file data/comprehensive_cars/images/*.jpg

• Here in this warning, for each time this message appears, the * stands for a different image filename from the dataset. Highly probable that none of the images could be loaded.

So, after following the CUDA-related message and using "the 'spawn' start method" by adding torch.multiprocessing.set_start_method('spawn') at the beginning of train.py, an exception is raised from this line:

  File "media/<user>/HD/HF_train/headnerf/train.py", line 14, in <module>
    torch.multiprocessing.set_start_method('spawn')
  File "/home/<user>/anaconda3/envs/test2/lib/python3.8/multiprocessing/context.py", line 243, in set_start_method
    raise RuntimeError('context has already been set')
RuntimeError: context has already been set

Can anyone tell what this means?
Or even if I'm on the right path to successfully run train.py?

Kind regards,
André

Hi, I am also using DECA. Have you found the way to estimate the extrinsic matrix (3 * 4) with DECA ? The code_dict['pose'] only returns six parameters. Thanks