US 12,411,886 B2
Systems and methods for searching audiovisual data using latent codes from generative networks and models
Ross F. Elliot, Parkland, FL (US); Seth Haberman, New York, NY (US); Michael A. Baumer, Bethesda, MD (US); and Nakul Dawra, Miami, FL (US)
Assigned to Unknot Inc., Parkland, FL (US)
Filed by Unknot Inc., Parkland, FL (US)
Filed on Feb. 1, 2023, as Appl. No. 18/163,114.
Application 18/163,114 is a continuation of application No. 17/093,386, filed on Nov. 9, 2020, granted, now 11,593,652.
Claims priority of provisional application 62/932,603, filed on Nov. 8, 2019.
Prior Publication US 2024/0184821 A1, Jun. 6, 2024
This patent is subject to a terminal disclaimer.
Int. Cl. G06F 16/53 (2019.01); G06F 16/532 (2019.01); G06F 16/535 (2019.01); G06F 18/2113 (2023.01); G06F 18/214 (2023.01); G06N 3/04 (2023.01); G06N 3/08 (2023.01); G06T 9/00 (2006.01); G06V 10/82 (2022.01); G06V 30/19 (2022.01); G06V 30/194 (2022.01)
CPC G06F 16/532 (2019.01) [G06F 16/535 (2019.01); G06F 18/2113 (2023.01); G06F 18/214 (2023.01); G06N 3/04 (2013.01); G06N 3/08 (2013.01); G06T 9/002 (2013.01); G06V 10/82 (2022.01); G06V 30/19173 (2022.01); G06V 30/194 (2022.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A system for searching source audiovisual data using a plurality of search queries, the system comprising:
a representative data set including a plurality of source audiovisual files;
a search feature space comprising a plurality of search feature codes;
a search module derived from the representative data set, configured to map a collection of source audiovisual files and a collection of search queries to a subset of the collection of source audiovisual files that satisfies the requirements specified by the collection of search queries, the search module comprising:
a trained generative model derived from the representative data set, the trained generative model comprising:
a plurality of latent spaces, each comprising a plurality of latent codes;
a plurality of trained generator mappings, each configured to map one or more latent codes to one or more generated audiovisual clips that share at least one characteristic feature with at least one of the source audiovisual files in the representative data set;
a generator-coupled compressor mapping configured to map one or more of the source audiovisual files to one or more resulting latent codes, wherein the plurality of trained generator mappings map the one or more resulting latent codes to one or more reconstructed audiovisual clips resembling the one or more audiovisual files;
one or more generator-coupled search feature identifiers which map latent codes to search feature codes, wherein nearby latent codes are mapped to nearby search feature codes; and
a filtering module configured to select latent codes from collections of latent code and search feature code pairs according to whether the search feature codes satisfy a subset of the collection of search queries.