US 12,266,156 B2
System and method for solving missing annotation object detection
Marios Savvides, Wexford, PA (US); Zhiqiang Shen, Pittsburgh, PA (US); Fangyi Chen, Pittsburgh, PA (US); and Han Zhang, Pittsburgh, PA (US)
Assigned to Carnegie Mellon University, Pittsburgh, PA (US)
Filed by CARNEGIE MELLON UNIVERSITY, Pittsburgh, PA (US)
Filed on Feb. 14, 2022, as Appl. No. 17/670,737.
Claims priority of provisional application 63/149,412, filed on Feb. 15, 2021.
Prior Publication US 2022/0262101 A1, Aug. 18, 2022
Int. Cl. G06V 10/774 (2022.01)
CPC G06V 10/774 (2022.01) 9 Claims
OG exemplary drawing
 
1. A method comprising:
exposing an object detection model to an input training image;
identifying one or more anchors on the image;
for each anchor:
determining a predicted confidence that the anchor is a ground-truth anchor;
applying a focal loss when the confidence is above a predetermined confusion threshold;
applying a mirror of the focal loss when the confidence is below the confusion threshold; and
generating a scaler value p as a confidence score for the anchor when the anchor has a predicted foreground object.