How to use BB-based filters

BB-based filters are those BB’s properties such as BB’s label, confidence, and IoU, etc. They apply to individual BBs, whereas sample-based filters apply to individual rows in the metrics/examples table. When using BB-based filters, the rows will not be filtered out until all of the ground truth BBs or predicted BBs are filtered out for those rows.

In the figure above, three BB-based filters, Confidence, IOU, and Label_predicted are applied. There are three orange boxes highlighting some numbers. In the Filters panel, (178/6,368) next to the Label head represents the numbers of filtered-in/total predicted BBs with the head label. In the chart, (1/3) represents the numbers of filtered-in/total predicted BBs on this image. In the Metrics panel, 123 is the number of filtered-in rows (samples) after applying these three filters.

Since no filters are applied to the ground truth BBs, all ground truth BBs are displayed on the images.

What do the numbers represent in filters?

For sample-based filters, such as Foreign Table, the numbers represent filtered-in/total rows if any filters are applied, or simply total rows if no filters are applied.

For BB-based filters, such as Label, the numbers represent filtered-in/total BBs if any filters are applied, or simply total BBs if no filters are applied.