Model Prediction Queries

Model Prediction Query Examples

“Show me all images with at least 1 geometric model prediction”

predictions.count > 0

As with annotations, category predictions cannot be queried using counts and won't be included in predictions.count

“Show me all images where my model predicted at least three pedestrians”

predictions.pedestrian >= 3

Model Prediction Query - Details

Unlike Annotations, where only one source of ground truth is supported, you can upload multiple Model Runs to the same Nucleus dataset. In the query bar, predictions always refers to the currently selected model run in the “Filter by Model Run” dropdown within Search Options. Querying based on predictions from multiple model runs at the same time is currently unsupported. Querying based on prediction-level metadata attributes is also currently unsupported.


`predictions.`	Prefix for all prediction-based queries
`predictions.count`	The total number of geometric predictions in the dataset item
`predictions.[class_name]`	The number of geometric `[class_name]` predictions in the dataset item
`predictions.[annotation_type].`	Prefix to filter based on prediction types. Useful to filter based on counts for a particular type of prediction, or if you have different prediction types (e.g., bounding boxes and instance segmentations) with identical class names, and you need to specify one of them in particular. `[annotation_type]` can be one of: `box`, `polygon`, `segmentation`, `category`
`predictions.[annotation_type].count`	The number of geometric predictions of type `[annotation_type]` in the dataset item
`predictions.[annotation_type].[class_name]`	The number of geometric `[class name]` predictions of type `[annotation_type]` in the dataset item
`predictions.[taxonomy_name]`	The `[class_name]` of the category prediction with taxonomy name `[taxonomy_name]` for the dataset item