Filtering your dataset
There are two different ways to set filters in Dataset Exploration:
- Set a default filter
- Set a custom filter
- Set a filter using via a chart on the metrics-tab
The default filters are available in the "Filter Dataset"-section. These include:
- Filter on items included in a specific chunk
- Filter on items that have been annotated only, predicted only, or both annotated and predicted
- Filter on items that have not been added to any chunk
By selecting an existing chunk from the dropdown labeled “Chunks”, you can create a filter that only shows the items that are part of that chunk.
In situations where you want to look at either:
- False negatives or positives in the annotations,
- False negatives or positives predicted by your model, or
- Classification errors
you can use the “Show items found in” dropdown and select one of the following options:
- Annotations only: Show items that have been annotated but not predicted. Useful e.g. when wanting to identify model false negatives.
- Predictions only: Show items that have been predicted but not annotated. Useful e.g. when wanting to identify annotation misses.
- Annotations OR Predictions: Show items that have been annotated or predicted. Useful when wanting to browse all items.
- Annotations AND Predictions: Show items that have been annotated and predicted. Useful when wanting to identify classification errors.
By ticking the checkbox "Hide items included in a chunk" you exclude all items that are added to any chunk.
Items filters are used to narrow your search by filtering on item attributes such as class, or a geometric attribute (e.g. size, distance from ego vehicle). Items filters can be created for either annotated or predicted items.
To create an item filter:
- Click the “Create filter” button in the “Filter Dataset”-section.
- Select items, either “Annotations” or “Predictions” in the first dropdown.
- Select a filter type, an operator (is or is not) and one or multiple values.
- Click the "Apply Filter"-button.
- Your filter is applied and added as a chip in the “Filter dataset” section.
