US 12,249,119 B2
Domain adaptation using pseudo-labelling and model certainty quantification for video data
Pavan Annangi, Bangalore (IN); and Deepa Anand, Bangalore (IN)
Assigned to GE PRECISION HEALTHCARE LLC, Waukesha, WI (US)
Filed by GE Precision Healthcare LLC, Wauwatosa, WI (US)
Filed on Mar. 8, 2022, as Appl. No. 17/654,019.
Claims priority of application No. 202141012577 (IN), filed on Mar. 23, 2021.
Prior Publication US 2022/0301287 A1, Sep. 22, 2022
Int. Cl. G06T 7/00 (2017.01); G06V 10/764 (2022.01); G06V 10/774 (2022.01)
CPC G06V 10/764 (2022.01) [G06T 7/0012 (2013.01); G06V 10/774 (2022.01); G06T 2207/20048 (2013.01); G06T 2207/20081 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method for domain adaptation, the method comprising:
obtaining using a same or different medical imaging devices a source data and a target data, wherein the source data and the target data comprises a plurality of frames for training a machine learning module;
testing the target data to identify if a minimum number of frames from the target data exhibit a frame confidence score based on the source data;
identifying at least one salient region within the target data and measuring a degree of spatial consistency of the at least one salient region over time;
identifying at least one class specific attention region within the frames of the target data and measuring a similarity score of the at least one class specific attention region within the frames of the target data as compared to a plurality of class-specific templates;
carrying out pseudo-labeling of the target data based on a module generated using source data and calculating a certainty metric value indicative of the accuracy of the pseudo-labeling on the target data;
wherein calculating the certainty metrics value comprises calculating using the frame confidence score, the degree of spatial consistency of the at least one salient region over time, the similarity score of the at least one class specific attention region within the frames of the target data as compared to the plurality of class-specific templates; and
retraining the machine learning module till the certainty metrics value reaches peak and further retraining the machine learning module does not increase the certainty metrics value.