How to succesfully run evaluation script ?

[orgicus]

Hi,

I've setup dependencies, downloaded and unpacked datasets and pretrained weights.

However, I'm running into errors running the eval script:

python scripts/evaluate.py -o output/pretrained_models/lab/
...
Traceback (most recent call last):
  File "scripts/evaluate.py", line 98, in <module>
    main(cfg)
  File "/root/miniforge3/envs/hugs/lib/python3.8/site-packages/torch/autograd/grad_mode.py", line 27, in decorate_context
    return func(*args, **kwargs)
  File "scripts/evaluate.py", line 69, in main
    trainer = GaussianTrainer(cfg)
  File "/mnt/e/DL/ml-hugs/./hugs/trainer/gs_trainer.py", line 212, in __init__
    betas = self.human_gs.betas.detach() if hasattr(self.human_gs, 'betas') else self.train_dataset.betas[0]
AttributeError: 'GaussianTrainer' object has no attribute 'train_dataset'

I've tweaked the code slightly, from:

if not cfg.eval:
            self.train_dataset = get_train_dataset(cfg)     

to: self.train_dataset = get_train_dataset(cfg)

However this leads to another missing attribute error:

Traceback (most recent call last):
  File "scripts/evaluate.py", line 98, in <module>
    main(cfg)
  File "/root/miniforge3/envs/hugs/lib/python3.8/site-packages/torch/autograd/grad_mode.py", line 27, in decorate_context
    return func(*args, **kwargs)
  File "scripts/evaluate.py", line 69, in main
    trainer = GaussianTrainer(cfg)
  File "/mnt/e/DL/ml-hugs/./hugs/trainer/gs_trainer.py", line 212, in __init__
    betas = self.human_gs.betas.detach() if hasattr(self.human_gs, 'betas') else self.train_dataset.betas[0]
AttributeError: 'NeumanDataset' object has no attribute 'betas'

I don't think tweaking the code to over the humps is the way forward.
What is the recommended way of running the evaluation script ?
Additionally, what is the recommended way of running inference and rendering results ?

Thank you for your time,
George

[GostInShell]

This is caused by an inconsitent variable name of cfg.human.name in the config file.
Change this line into the following will solve it:
elif cfg.human.name == 'hugs_trimlp' or cfg.human.name == 'hugs_triplane':

[orgicus]

@GostInShell Tank you so much! That solved the issue !

Unfortunately when I ran the script it seems betas is missing from the dataset ?

Traceback (most recent call last):
  File "scripts/evaluate.py", line 98, in <module>
    main(cfg)
  File "/root/miniforge3/envs/hugs/lib/python3.8/site-packages/torch/autograd/grad_mode.py", line 27, in decorate_context
    return func(*args, **kwargs)
  File "scripts/evaluate.py", line 69, in main
    trainer = GaussianTrainer(cfg)
  File "/mnt/e/DL/ml-hugs/./hugs/trainer/gs_trainer.py", line 219, in __init__
    self.static_smpl_params = get_smpl_static_params(
  File "/mnt/e/DL/ml-hugs/./hugs/datasets/utils.py", line 150, in get_smpl_static_params
    if betas.shape != (10,):
AttributeError: 'NoneType' object has no attribute 'shape'