Large Vocabulary Instance Segmentation (LVIS)
I Know It When I See It: Zero-shot detection for objects not in training data.
Object detectors typically detect only items that were labeled in their training data. A new method liberates them to locate and recognize a much wider variety of objects.