Amazon Technologies, Inc.

Rohith Mysore Vijaya Kumar

Amit Kumar Agrawal

Ambrish Tyagi

Yadunandana Nagaraja Rao

Apoorv Chaudhri

Abhishek Singh

Visesh Uday Kumar Chari

Suresh Bholabhai Lakhani

Matthew Alan Townsend

Prakash Ramu

Eduard Oks

Mashhour Solh

Qiang Liu

Lelin Zhang

Siyi Ding

Xiaoran Wang

Sunil Sharadchandra Hadap

Brandon Michael Smith

Jinjin Li

Ram Sever

David Ting-Yu Wu

Rajesh Gautam

Dylan John Drover

Noam Sorek

Paul Aksenti Savastinuk

Vimal Bhat

Weijian Wang

Najmeh Sadoughi Nourabadi

Daniel Stephen ReMine

Sharath Kumar Gopal

Abhishek Yanamandra

Techniques for performing a machine learning (ML) cinematic (e.g., movie) question answering are described. According to some examples, a computer-implemented method includes receiving a request from a viewer device at a content delivery service to play a video; sending the video from the content delivery service to the viewer device; receiving, by the content delivery service, a question from the viewer device during playing of the video; generating, by a script context retrieval machine learning model of the content delivery service, a proper subset of a script of the video based on an input of the question; generating, by a cinematic question answering machine learning model of the content delivery service, an answer based on an input of the proper subset of the script of the video; and sending, by the content delivery service, the answer to the viewer device.

Computer-implemented methods for movie question answering as a second screen experience using a large language machine learning model

Technologies are disclosed for managing composite storefront images. The composite storefront images can be generated utilizing portions of media content (e.g., visual content, such as videos, trailers, etc.), and templates generated based on sets of existing storefront images. Objects can be extracted from frames and composited together to generate course composite images. The course composite storefront images can be utilized to map the object portions onto the media content artwork styles to generate refined composite storefront images based on the templates.

Composite storefront image management

Described are systems and methods directed to generation of a personalized three-dimensional ('3D') body model of a body, such as a human body, based on two-dimensional ('2D') images of that body and the generation and presentation of predicted personalized 3D body models of the body when one or more body measurements (e.g., body fat, body weight, muscle mass) are changed. For example, a user may provide a target body measurement value and the implementations will generate one or more predicted personalized 3D body models representative of a predicted appearance of the body with the target body measurement value.

Generation and presentation of predicted personalized three-dimensional body models

Devices, systems and methods are disclosed for improving story assembly and video summarization. For example, video clips may be received and a theme may be determined from the received video clips based on annotation data or other characteristics of the received video data. Individual moments may be extracted from the video clips, based on the selected theme and the annotation data. The moments may be ranked based on a priority metric corresponding to content determined to be desirable for purposes of video summarization. Select moments may be chosen based on the priority metric and a structure may be determined based on the selected theme. Finally, a video summarization may be generated using the selected theme and the structure, the video summarization including the select moments.

Video ingestion and clip creation

This application describes techniques for providing computer vision for manual services. In some instances, a remote system may determine that a current time is within a threshold period of time of a scheduled service and, based on the determination, send a first message requesting image data to a camera apparatus located within an environment. After sending the first message, the remote system may receive image data associated with the environment from the camera apparatus and use the image to detect an object within the environment. The remote system can then send a second message to a device of a guest and a third message to a device of a user, where each of the second message and the third message indicates that the object is within the environment. In some instances, the object can include an animal while in some instances, the object can include a person.

Computer vision to enable services

Predictive personalized three-dimensional body models

A system and method for generating preview data from video data and using the preview data to select portions of the video data or determine an order with which to upload the video data. The system may sample video data to generate sampled video data and may identify portions of the sampled video data having complexity metrics exceeding a threshold. The system may upload a first portion of the video data corresponding to the identified portions while omitting a second portion of the video data. The system may determine an order with which to upload portions of the video data based on a complexity of the video data. Therefore, portions of the video data that may require additional processing after being uploaded may be prioritized and uploaded first. As a result, a latency between the video data being uploaded and a video summarization being received is reduced.

Preview streaming of video data

Devices, systems and methods are disclosed for reducing a perceived latency associated with uploading and annotating video data. For example, video data may be divided into video sections that are uploaded individually so that the video sections may be annotated as they are received. This reduces a latency associated with the annotation process, as a portion of the video data is annotated before an entirety of the video data is uploaded. In addition, the annotation data may be used to generate a master clip table and extract individual video clips from the video data.

A system and method for selecting portions of video data from preview video data is provided. The system may extract image features from the preview video data and discard video frames associated with poor image quality based on the image features. The system may determine similarity scores between individual video frames and corresponding transition costs and may identify transition points in the preview video data based on the similarity scores and/or transition costs. The system may select portions of the video data for further processing based on the transition points and the image features. By selecting portions of the video data, the system may reduce a bandwidth consumption, processing burden and/or latency associated with uploading the video data or performing further processing.

Frame selection of video data

This application describes techniques for providing computer vision-based surveillance of in-home services. In some instances, a remote system may authenticate a guest at an environment using data that the remote system receives from a camera apparatus. To authenticate the guest, the remote system can determine that an identification of the guest matches an identification of a worker associated with a scheduled service at the environment. When the scheduled service includes an in-home delivery, the remote system can further determine that a package in possession of the guest matches an item for the scheduled in-home delivery. In some instances, after the guest enters the environment, the remote system can further monitor the environment to ensure that the guest stays within a designated location of the environment. The remote system can further send a message to an electronic device when the guest is outside of the designated location.

Computer vision-based surveillance for services

Techniques are generally described for discontinuity detection. A first and second frame of video data representing a portion of a physical environment are received. The first and second frames are divided into first and second blocks of pixels, respectively. A set of matching blocks are determined by matching each block of the first blocks with a corresponding block of the second blocks. A difference value is determined for each pair of matching blocks of the set of matching blocks. A mean difference value for the set of matching blocks. The mean difference value is convolved with a convolution kernel. A determination is made that a result of the convolving the mean difference value with the convolution kernel exceeds a mean threshold value. A determination is made that a discontinuity between the first frame and the second frame is not due to motion in the physical environment.

Discontinuity detection in video data

Devices and techniques are generally described for image editing. In various examples, color image data and segmentation image data may be identified. In some examples, the segmentation image data may identify a first portion of the color image data as a foreground region and a second portion of the color image data as a background region. A segmentation boundary may be identified between the foreground region and the background region. A first area of the second portion may be identified, where the first area extends from the segmentation boundary to a first area outer boundary. A second area of the second portion may be identified, where the second area extends from the first area outer boundary to a second area outer boundary. A blended and blurred representation of the first area and the second area may be generated.

Growing community of inventors

Sunnyvale, CA, United States of America

Rohith Mysore Vijaya Kumar