Is there an existing issue for this?

• I have searched the existing issues

Bug description

I wanted to run triangulation, following training and analysis with the pytorch engine.
I made sure that pytorch was specified as the engine in the config.yaml files used for both the 3D and the original 2D analysis.
I encountered errors that suggested that perhaps the triangulation function is not prepared to process pytorch output.

I see the error:
"FileNotFoundError: It seems the model for iteration 0 and shuffle 1 and trainFraction 0.95 does not exist."
from line 49 of deeplabcut\pose_estimation_tensorflow\config.py.

Further error output suggests that the relevant "open()" call is seeking pose_cfg.yaml in a nested path underneath "dlc-models", when my pose_cfg.yaml files instead appear under "dlc-models-pytorch".

Is it correct to interpret this as saying that the triangulate() function does not yet support pytorch output?

In the meantime, I was assuming that I would either try to re-run training using tensorflow, or create a personal modified copy of triangulate() that would work around the problematic lines (since I was planning to use already-analyzed output.) I would welcome any suggestions for any easier fixes, though.

Operating System

Windows 10

DeepLabCut version

3.0.0rc3

DeepLabCut mode

single animal

Device type

NVIDIA Quadro RTX 4000

Steps To Reproduce

At the anaconda command prompt, I ran "deeplabcut.triangulate(config_path3d,videopath, filterpredictions=False)", where config_path3d pointed to the attached config file (uploaded within a zip file), and videopath pointed to the directory housing the videos. That ultimately produced the error I described above.

config.zip

Relevant log output

No response

Anything else?

No response

Code of Conduct

• I agree to follow this project's Code of Conduct