US 11,948,387 B2
Optimized policy-based active learning for content detection
Sumit Shekhar, Bengaluru (IN); Bhanu Prakash Reddy Guda, Podili (IN); Ashutosh Chaubey, Chhattisgarh (IN); Ishan Jindal, Mansa (IN); and Avneet Jain, Madhya Pradesh (IN)
Assigned to ADOBE INC., San Jose, CA (US)
Filed by ADOBE INC., San Jose, CA (US)
Filed on Feb. 8, 2021, as Appl. No. 17/170,307.
Prior Publication US 2022/0253630 A1, Aug. 11, 2022
Int. Cl. G06V 40/10 (2022.01); G06F 18/21 (2023.01); G06F 18/211 (2023.01); G06F 18/214 (2023.01); G06N 20/00 (2019.01); G06V 20/20 (2022.01)
CPC G06V 40/10 (2022.01) [G06F 18/211 (2023.01); G06F 18/2155 (2023.01); G06F 18/2178 (2023.01); G06N 20/00 (2019.01); G06V 20/20 (2022.01)] 17 Claims
OG exemplary drawing
 
1. A method for training an object detection network, the method comprising:
training an object detection network using a labeled training set, wherein each element of the labeled training set includes an image and ground truth labels for object instances in the image;
predicting annotation data for a candidate set of unlabeled data using the object detection network;
selecting a sample image from the candidate set using a policy network;
generating a labeled sample based on the selected sample image and the annotation data, wherein the labeled sample includes labels for a plurality of object instances in the sample image;
performing additional training on the object detection network based at least in part on the labeled sample; and
predicting additional annotation data for a state set, wherein the sample image is selected based at least in part on the additional annotation data.