ValueError: not enough values to unpack (expected 2, got 0)

[cmj12345]

Hello, when I run "bash scripts/train_bunny_real.sh", after I close the picture in Open3D, then when trainning, it reported an error.

Traceback (most recent call last):
File "/home/chengmj/anaconda3/envs/touch-gs/bin/ns-train", line 8, in
sys.exit(entrypoint())
File "/home/chengmj/anaconda3/envs/touch-gs/lib/python3.8/site-packages/nerfstudio/scripts/train.py", line 263, in entrypoint
tyro.cli(
File "/home/chengmj/anaconda3/envs/touch-gs/lib/python3.8/site-packages/tyro/_cli.py", line 169, in cli
output = _cli_impl(
File "/home/chengmj/anaconda3/envs/touch-gs/lib/python3.8/site-packages/tyro/_cli.py", line 373, in _cli_impl
parser_spec = _parsers.ParserSpecification.from_callable_or_type(
File "/home/chengmj/anaconda3/envs/touch-gs/lib/python3.8/site-packages/tyro/_parsers.py", line 125, in from_callable_or_type
field_out = handle_field(
File "/home/chengmj/anaconda3/envs/touch-gs/lib/python3.8/site-packages/tyro/_parsers.py", line 368, in handle_field
subparsers_attempt = SubparsersSpecification.from_field(
File "/home/chengmj/anaconda3/envs/touch-gs/lib/python3.8/site-packages/tyro/_parsers.py", line 578, in from_field
subparser = ParserSpecification.from_callable_or_type(
File "/home/chengmj/anaconda3/envs/touch-gs/lib/python3.8/site-packages/tyro/_parsers.py", line 125, in from_callable_or_type
field_out = handle_field(
File "/home/chengmj/anaconda3/envs/touch-gs/lib/python3.8/site-packages/tyro/_parsers.py", line 385, in handle_field
if force_primitive == "struct" or _fields.is_struct_type(
File "/home/chengmj/anaconda3/envs/touch-gs/lib/python3.8/site-packages/tyro/_unsafe_cache.py", line 33, in wrapped_f
out = f(*args, **kwargs)
File "/home/chengmj/anaconda3/envs/touch-gs/lib/python3.8/site-packages/tyro/_fields.py", line 208, in is_struct_type
list_or_error = field_list_from_type_or_callable(
File "/home/chengmj/anaconda3/envs/touch-gs/lib/python3.8/site-packages/tyro/_fields.py", line 245, in field_list_from_type_or_callable
registry.get_primitive_spec(PrimitiveTypeInfo.make(f, set()))
File "/home/chengmj/anaconda3/envs/touch-gs/lib/python3.8/site-packages/tyro/constructors/_registry.py", line 146, in get_primitive_spec
maybe_spec = spec_factory(type_info)
File "/home/chengmj/anaconda3/envs/touch-gs/lib/python3.8/site-packages/tyro/constructors/_primitive_spec.py", line 472, in dict_rule
key_type, val_type = get_args(type_info.type)

ValueError: not enough values to unpack (expected 2, got 0)

And in addition, I wonder if "bash scripts/train_bunny_blender.sh" can run successfully, because I see that there is already relevant code, but it is still written in ReadMe as ' (processing to a unified format in progress)'.

Thank you very much!!!!

[peasant98]

Hi @cmj12345 , thanks for checking out our work!

This is an interesting issue that I haven't seen before, but it seems to come down to an issue in the ns-train command.

ns-train depth-gaussian-splatting --data bunny_blender_data/ --viewer.quit-on-train-completion True --pipeline.model.depth-loss-mult 0.5 --pipeline.model.depth-loss-type SIMPLE_LOSS nerfstudio-data --train-split-fraction 0.13

Two thoughts I have:

1. Are you using our forked version of nerfstudio in this repo? It is a bit older than the current version available on Github.

2. Maybe there is a version issue for tyro.

Let me know how it goes!

[cmj12345]

Yes, I use your forked version of nerfstudio in this repo.

And I found a warning before training. Perhaps it's an abnormal situation?

Fusing vision and touch...
utils/fuse_touch_vision.py:117: RuntimeWarning: divide by zero encountered in divide
  uncertainty_touch_valid_reciprocal = 1 / uncertainty_touch_valid
utils/fuse_touch_vision.py:142: RuntimeWarning: invalid value encountered in divide
  mu_touch_var_weighted = mu_touch_valid / uncertainty_touch_valid

And if it's normal, can you provide recommended version for tyro?

Thanks a lot!

[peasant98]

Hmm...

So tyro should be at a version greater than 0.5.10.

ns-train depth-gaussian-splatting --data bunny_blender_data/ --viewer.quit-on-train-completion True --pipeline.model.depth-loss-mult 0.5 --pipeline.model.depth-loss-type SIMPLE_LOSS nerfstudio-data --train-split-fraction 0.13

What is the output of this as well?

I also notice this, do you think it could be related?

[cmj12345]

I solved it. It's a version issue of tyro. I downgraded it.
Thanks a lot !