US 11,881,022 B2
Weakly-supervised action localization by sparse temporal pooling network
Ting Liu, Playa Vista, CA (US); Gautam Prasad, Los Angeles, CA (US); Phuc Xuan Nguyen, San Diego, CA (US); and Bohyung Han, Seoul (KR)
Assigned to GOOGLE LLC, Mountain View, CA (US)
Filed by Google LLC, Mountain View, CA (US)
Filed on Mar. 10, 2023, as Appl. No. 18/181,806.
Application 18/181,806 is a continuation of application No. 16/625,172, granted, now 11,640,710, previously published as PCT/US2018/059176, filed on Nov. 5, 2018.
Claims priority of provisional application 62/586,078, filed on Nov. 14, 2017.
Prior Publication US 2023/0215169 A1, Jul. 6, 2023
This patent is subject to a terminal disclaimer.
Int. Cl. G06V 20/40 (2022.01); G06F 18/214 (2023.01); G06F 18/243 (2023.01)
CPC G06V 20/40 (2022.01) [G06F 18/214 (2023.01); G06F 18/24317 (2023.01); G06V 20/44 (2022.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method for temporally localizing a target action in a video, comprising:
inputting a video into a machine-learned model comprising one or more weakly supervised temporal action localization models;
analyzing the video by the one or more weakly-supervised temporal action localization models to determine one or more weighted temporal class activation maps; and
determining a temporal location of a target action in the video based at least in part on the one or more weighted temporal class activation maps;
wherein the machine-learned model comprises a sparse temporal pooling network comprising a first weakly supervised temporal action localization model and a second weakly supervised temporal action localization model.