Amazon Technologies, Inc.

Aneeb Qureshi

Samuel Scott Gigliotti

Prakash Ramu

David James Roy Erdmann

Ita Lifshitz

David Lawrence Seymore

Megan Heffernan

Adam Paul DeRusha

Anta Imata Safo

Srikanth Thirumalai

Robert Yuji Haitani

Amber Autrey Taylor

Dan Catalin Teodorescu

Jean Joseph Tavernier

Xuping Zhang

Karthik Krishnamurthy

Jason Juin-Yip Kwok

Brian Elieson

Asia Noel VanderBilt

Maher Alam Beg

Joshua Gene Slider

Chee H Chew

Cory William Toedebusch

Songqing Cai

Described are systems and method directed to generation of a dimensionally accurate three-dimensional ('3D') body model of a body, such as a human body, based on two-dimensional ('2D') images of that body. A user may use a 2D camera, such as a digital camera typically included in many of today's portable devices (e.g., cell phones, tablets, laptops, etc.) and obtain a series of 2D body images of their body from different directions with respect to the camera. The 2D body images may then be used to generate a plurality of predicted body parameters corresponding to the body represented in the 2D body images. Those predicted body parameters may then be further processed to generate a dimensionally accurate 3D model of the body of the user.

Two-dimensional image collection for three-dimensional body composition modeling

Techniques are described for allocation hardware and/or bandwidth resources of a viewer device between a live streaming media content playback of a video player and a concurrently provided live chat. For example, the network and hardware resources used by a video player for streaming media content playback can be adjusted based on chat conditions of the live chat.

Resource management for video playback and chat

Techniques for improving access to and interactions with bots are described. In an example, a first bot, hosted on a computing system, may identify an action to be performed for a user associated with a computing device. The action may be identified based on a user interaction with the first bot, where the user interaction may be provided from the computing device. The first bot may select a second bot based on the action. The second bot may be hosted on a same or a different computing system. The first bot may determine, based on a previous user input to the first bot, a parameter value for an execution of the action and may send the parameter value to the second bot. In response, the first bot may receive a result of an execution of the action and may provide information about the result to the computing device.

Bot search and dispatch engine

Recommendations for media assets based on product similarities are provided. In some embodiments, a recommendation can be determined using asset-to-asset similarity information that can be obtained by mapping media assets to products and further mapping products to media assets based at least on the prior mapping. In addition, product similarity information can be leveraged to generate groups of media assets for different categories of similarity present in the product similarity information. Such groups can be identified, for example, by using the mapped products to media assets as applied to the different categories. Further a diverse arrangement of media assets can be generated by selecting media assets from the groups according to a defined selection rule.

Process and machine for media determination using product information

A machine learning engine may correlate user profile data and/or list item data with a service provider action that may lead to a particular user action with respect to a list item (e.g., a task item of a to-do list, a content item of a content item queue, etc.). For example, a task item performance application may process user profile data and/or task item profile data to generate training data. In addition, the machine learning engine may generate a task item performance model using the training data by identifying indicators that correlate with the particular user action. Additionally, the task item performance application may then use the task item performance model to suggest actions that may be performed by a service provider to increase the likelihood of the user performing the particular user action with respect to the list item. Further, the machine learning engine may also determine weights for individual indicators.

Machine learning based list recommendations

Disclosed are various embodiments for offline advertisement presentation. Media items and advertisements are stored on a client device. During playback of the media item, in absence of a network connection, a stored advertisement is presented to a user. This may be facilitated by a proxy advertising service implemented in the client device.

Offline video advertising

Disclosed are various embodiments for a media player customization application. A media player framework defines basic functionality to render media items. Plugins may be added to the media player framework to implement desired functionality. The plugin model allows for a media player to minimize file sizes by only included required functionality.

Customizable media player framework

Described herein are a system and methods for providing relevant search results to a user using an automated user assistant to update filters used as search parameters. In some embodiments, an automated user assistant may initiate a conversation with a user detected interacting with an item. The user may provide feedback that indicates one or more attributes relevant to the user's interests with respect to the currently viewed item. In some embodiments, the user may also provide an indication of how values associated with those attributes would be made more relevant to the user. Various filters associated with the attributes may be updated based on the received user feedback. The updated filters may then be provided as search parameters to a search engine. A set of search results returned by the search engine may be provided to the user.

Conversational bot to navigate upwards in the funnel

Disclosed are various embodiments analyzing a user's interaction with his or her music library. The system generates a time series by tracking a plurality of instances of music library interaction between a user and a music library. The system also determines a distribution expressed in the time series, the distribution indicating a trend of playing a set of audio items for a particular period of time, the set of audio items being associated with a class, wherein a timestamp corresponds to an occurrence of the trend. The system associates the distribution with a triggering event and generates a recommendation according to the class in response to the triggering event.

Media library analyzer

Techniques are described for allocation hardware and/or bandwidth resources of a viewer device between a live streaming media content playback of a video player and a concurrently provided live chat. For example, the bandwidth used by the live chat can be adjusted based on the playback of the video player.

A localization service may be provided to manage distribution of content including text overlays according to language codes. In some examples, the language codes include regional identifiers and language identifiers. The localization service may provide the content including translated text overlays for reproduction by client devices in accordance with descriptive data associated with the images. In some examples, the descriptive data may describe the text overlays, placement of the text overlays with respect to the content, and characteristics of the text overlays.

Content localization

Techniques are described for providing rendition manifests for the playback of video content or other types of content on a user device. The response to a request for content may serially include data objects for one or more rendition manifests as Data Uniform Resource Indicators (URIs) or in another data format. By incorporating the data of the rendition manifest(s) in the response, implementations may forego a network request to retrieve the rendition manifest(s) and thus reduce the latency for presenting content. In some cases, the incorporated rendition manifest(s) may list locations for a subset of chunks of a content file, and the content may be marked as live content, to reduce the size of the response and further reduce latency. In some cases, locations for chunks from multiple content files may be included in a same set of one or more rendition manifests in the response.

Growing community of inventors

Seattle, WA, United States of America

Aneeb Qureshi