Closed BENULL closed 1 year ago
Dear both, thank you for your interest.
LOST by itself is designed to discover a single object per image. But we additionally propose to train a simple class-agnostic detector (named CAD) with LOST predictions as pseudo-boxes which at inference produce multi-objects predictions. Please see the section Towards unsupervised object detection
Hope that helps !
I have the same question. Have you find out?