Operator Reference
fit_dl_out_of_distribution (Operator)
fit_dl_out_of_distribution — Extend a deep learning model for Out-of-Distribution Detection.
Signature
fit_dl_out_of_distribution( : : DLModelHandle, DLDataset, GenParam : )
Description
fit_dl_out_of_distribution extends a trained deep learning model
DLModelHandle of 'type' = 'classification'
for Out-of-Distribution Detection. This functionality allows the model to
detect samples that
differ significantly from the classes it was trained on, known as
“Out-of-Distribution” (OOD) samples.
When apply_dl_model is called subsequently, the results will include
the following additional entries related to Out-of-Distribution Detection:
- 'ood_result':
Indicates whether the sample is predicted as out-of-distribution.
- 'ood_score':
Indicates how much the sample differs from the trained classes. The higher this score, the more likely it is that the sample is out-of-distribution.
- 'ood_threshold':
If 'ood_score' exceeds this threshold, the sample is predicted as out-of-distribution. The out-of-distribution threshold is computed during the execution of
fit_dl_out_of_distributionand stored within the model handleDLModelHandleas'ood_threshold'. If required, the threshold can be adjusted manually using the operatorset_dl_model_param.
For fit_dl_out_of_distribution to work properly, it
is important that DLDataset is the same dataset with the same split
and preprocessing parameters, as the one used for training
DLModelHandle.
It is crucial that the provided dataset DLDataset contains diverse
and sufficient samples for each class to ensure reliable
Out-of-Distribution Detection. If the dataset is too small or lacks variation,
fit_dl_out_of_distribution may return an error.
In such cases, additional training data should be added to the dataset.
fit_dl_out_of_distribution can be applied to any classification model
supported by HALCON. For models created using
Deep Learning / Framework operators or read from an
ONNX model file, Out-of-Distribution Detection compatibility may
vary depending on the architecture.
The performance of the model for Out-of-Distribution Detection can be
evaluated using the procedure evaluate_dl_model.
To evaluate the model on out-of-distribution data, these can be added
to the DLDataset using
the procedure add_dl_out_of_distribution_data,
allowing for testing whether the model can accurately separate
in-distribution from out-of-distribution data.
Adjustments to the 'ood_threshold' will affect evaluation results.
Therefore, it is recommended to re-evaluate the model after making
such changes.
GenParam is a dictionary for setting generic parameters.
Currently no generic parameters are supported.
Attention
If fit_dl_out_of_distribution is called for a model that has already
been extended with Out-of-Distribution Detection, the previous internal
calculations are discarded and the model is adapted anew
Certain modifications to the model, such as changing the number of classes
or continuing training of the model,
cannot be performed once the model has been extended for
Out-of-Distribution Detection.
To make such changes possible, the model internal Out-of-Distribution Detection
must first be removed from the model using the parameter
'clear_ood' in set_dl_model_param.
Once removed, fit_dl_out_of_distribution
can be called again to re-enable Out-of-Distribution Detection.
Execution Information
- Multithreading type: reentrant (runs in parallel with non-exclusive operators).
- Multithreading scope: global (may be called from any thread).
- Processed without parallelization.
Parameters
DLModelHandle (input_control) dl_model → (handle)
Handle of a deep learning classification model.
DLDataset (input_control) dict → (handle)
Dataset, which was used for training the model.
GenParam (input_control) dict → (handle)
Dictionary for generic parameters.
Possible Predecessors
Module
Deep Learning Professional