OOD Detection using COOD #493
Conversation
|
Keeping it as Draft. TODO : Notebook is not fully there. Needs some comments and descriptions |
There was a problem hiding this comment.
Hey Rajesh, many thanks for the PR. I haven't looked at the notebook yet, the rest of the pr looks good though. I do feel that the 'COODModel' class is pretty cluttered, I think it would be better to split this up into two classes: The 'COODModel' which handles the data preprocessing and prediction, and a new 'OODTrainer' class that interacts with Geti, creates the deployment, pulls the data and prepares it for training. The 'OODTrainer' could output the 'COODModel' as a result of its 'train' method. This would make the feature more modular, rather than putting all the responsibilities in a single class. What do you think?
geti_sdk/detect_ood/ood_model.py
Outdated
| self._thresholds_prefix + threshold_name | ||
| ] | ||
|
|
||
| def _train_cood_hpo( |
There was a problem hiding this comment.
It looks like this is not used at the moment?
Need optimisatio, better training code,
Yet to check/test the inference and training flow completely.
rajeshgangireddy
force-pushed
the
rgangire/ood_detection_trigger
branch
from
5937026
to
8f2af0a
Compare
This PR adds :