API Design Choice Behind Setting Prediction Threshold on The Model

[wescef-fahmad]

One strange issue with Sahi is that model creation APIs like AutoDetectionModel.from_pretrained() accept a prediction threshold parameter, like confidence_threshold but the prediction APIs, for example get_sliced_prediction() do not allow setting/overriding a prediction threshold. This makes it difficult to use the API in setting where the threshold needs to be varied dynamically.

Is this a deliberate design choice?