Systems and methods for action-based split-screen video generation
OwnerJean-Yves Couleaud, Evgeny Kaminsky +3inventors
InventorsJean-Yves Couleaud, Evgeny Kaminsky, Charles Dasher, Tao Chen, Ning Xu
Life of this patent
Estimating the term…
Abstract
Methods and systems are described herein for efficient video-to-video search and composite video generation. In an example system, the system receives a video, via a user device, and extracts action embeddings and pose embeddings. The system identifies a first subset of videos based on the extracted action embeddings. The system identifies a second subset of videos from the first subset of videos based on the extracted pose embeddings. The system receives a selection of a video from the second subset of videos. The system generates for display a composite video comprising the first video and the selected video.
Claims21
A method comprising:
- receiving a first video via a user device;
- extracting, from the first video, at least one action embedding;
- extracting, from the first video, at least one pose embedding;
- identifying a first subset of videos from a video database based at least in part on the extracted at least one action embedding of the first video;
- identifying a second subset of videos from the first subset of videos based at least in part on the extracted at least one pose embedding of the first video;
- receiving, via a user interface of the user device, selection of a video from the second subset of videos;
- and
- generating for display a composite video comprising at least part of the first video and at least part of the selected video.
A system comprising:
- memory;
- control circuitry configured to:
- store a first video in the memory;
- extract, from the first video, at least one action embedding;
- extract, from the first video, at least one pose embedding;
- identify a first subset of videos from a video database based at least in part on the extracted at least one action embedding of the first video;
- identify a second subset of videos from the first subset of videos based at least in part on the extracted at least one pose embedding of the first video;
- receive, via a user interface, selection of a video from the second subset of videos;
- and
- cause to provide for display a composite video comprising at least part of the first video and at least part of the selected video.
50. (canceled)
Dependencies are inferred from the claim text.