Pytorch not handeling missing body parts well. #3114
Description
Is there an existing issue for this?
- I have searched the existing issues
Operating System
Windows 11
DeepLabCut version
3.0.0rc13
What engine are you using?
pytorch
DeepLabCut mode
single animal
Device type
RTX 4070
Bug description 🐛
I recently switched from Tensorflow to Pytorch. When I trained a new pytorch model with the same dataset, the output is really poor with a p-cutoff of 0.6. The evaluation output looks similar, just really low likelihood.
The original Tensorflow model was trained with an original dataset, then several more trainings with added labelled outlier frames. The outlier frames mostly had invisible body parts so I had to remove the annotaion of those body parts.
The pytorch model I trained gives me extremely low p value when analyzing videos even in frames where the body part is visible. I campared to Tensorflow and the likelihood value wend down from 0.99 to 0.30 between the models.
Does the two models handel missing body parts differently? I am not sure if I should remove some labelled frames and train on a new dataset.
Steps To Reproduce
No response
Relevant log output
Anything else?
No response
Code of Conduct
- I agree to follow this project's Code of Conduct