Exclusion polygons / repeat detection suppression

[nathanielrindlaub]

I've come across a few different approaches for dealing with rocks/branches/other background features that are causing repeated false-positives:

1. Allow users to select the offending bounding box of a recurring false positive detection and suppress/invalidate all future (and/or) prior predictions with very similar bounding boxes.
2. Allow users to draw polygons on their images that will suppress/invalidate all future/prior predictions within them (Morgan's idea)
3. Semi-automated flagging and filtering of "suspicious" detections, similar to what Dan implemented in his repeat_detection_elimination scripts.

This is probably an incomplete list, but some combination of them is worth considering and potentially implementing.