Abstract: A model generation system generates three-dimensional object models based on two-dimensional images of an object. The model generation system can apply an iterative gradient decent process to model parameters for part models within an object model to compute a final set of model parameter values to generate the object model. To compute the final set of model parameter values, the model generation generates a reference image of the object model and compares the reference image to a received image. The model generation system uses a differentiable error function to score the reference image based on a received image. The model generation system updates the set of model parameter values based on the score for the reference image, and iteratively repeats the process until a reference image is sufficiently similar to the received image.

Abstract: A model generation system generates three-dimensional object models based on two-dimensional images of an object. The model generation system can apply an iterative gradient decent process to model parameters for part models within an object model to compute a final set of model parameter values to generate the object model. To compute the final set of model parameter values, the model generation generates a reference image of the object model and compares the reference image to a received image. The model generation system uses a differentiable error function to score the reference image based on a received image. The model generation system updates the set of model parameter values based on the score for the reference image, and iteratively repeats the process until a reference image is sufficiently similar to the received image.

Abstract: The model generation system may generate a 3D object model based on computer aided design (CAD) data describing an object. The CAD data received by the CAD conversion module 160 may contain a set of surfaces for the object. Each surface may be described by a surface equation that describes the shape of the surface in a 3D space. The model generation system may extract those surface equations from the CAD data and generate field lines and equipotential lines. The field lines may be lines that are tangent to the gradient vector field of the surface, and the equipotential lines may be lines along the surface that designate points that have the same potential within the gradient field vector. The model generation system may use the field lines and the equipotential lines to generate quadrangular tessellations for a 3D object model for the object described by the CAD data.

Abstract: A computer-implemented method of managing questions and answers on a computer-hosted service. The method includes a computing device receiving text based tax question and answer pairings and inputting the tax question and answer pairings into a content model executed by the device and outputting a content score for each tax question and answer pairing based on the model. The content score comprises a number within a range. One end of the range corresponds to product content and another end of the range corresponds to general tax content. The device outputs an answer quality score for the tax question and answer pairings based at least in part on the content score and votes assigned to each respective question and answer pair, wherein votes comprises up votes and down votes. The device may generate a FAQ list stored in a database based at least in part on the answer quality score.

Abstract: A model generation system generates three-dimensional (3D) models for objects based on two-dimensional (2D) images of the objects. The model generation system may receive object images and generate a 3D object model for the object based on the object image. The model generation system may generate an object skeleton for the object based on the object image. The model generation system may use the object skeleton to generate pixel partitions representing parallel cross sections of the object. The model generation system may apply a machine-learning model (e.g., a neural network) to the object image to determine parameters for a shape that would best represent each parallel cross section and then generate the 3D object model for the object based on the shapes of each cross section, the object image, and the object skeleton.

Abstract: The model generation system may receive an object image and identify the parts of the object based on the object image. The model generation system may determine whether the model generation system has stored a part model that would correspond to the identified part. The model generation system may compare the portion of the object image that corresponds to the part to model images of a set of part models. The model generation system may identify a part model that best corresponds to the identified part based on similarity scores of the model images associated with the part model and the portion of the object image associated with the identified part. The model generation system may perform this process for each part of the object and then assemble an object model based on the part models for each part of the object.

Abstract: The model generation system may generate texture maps for a texture based on a material image. A material image is an image of a physical material that describes the color (e.g., red-green-blue (RGB) color system model) of the physical material. The model generation system may identify a material class for the physical material depicted in the material image by applying a machine-learning model to the material image. The model generation system may then identify a texture map model that generates texture maps for the physical material based on the material image. The texture map model is a machine-learning model that is trained to generate texture maps for material images of a particular material class. The texture maps generated by the texture map model may include texture maps of standard texture values, such as metalness and roughness.

Abstract: The model generation system may generate a 3D object model based on computer aided design (CAD) data describing an object. The CAD data received by the CAD conversion module 160 may contain a set of surfaces for the object. Each surface may be described by a surface equation that describes the shape of the surface in a 3D space. The model generation system may extract those surface equations from the CAD data and generate field lines and equipotential lines. The field lines may be lines that are tangent to the gradient vector field of the surface, and the equipotential lines may be lines along the surface that designate points that have the same potential within the gradient field vector. The model generation system may use the field lines and the equipotential lines to generate quadrangular tessellations for a 3D object model for the object described by the CAD data.

Abstract: A computer-implemented method of generating rich content webpages from a question and answer (Q&A) library includes providing a topic and one or more seed questions related to the topic. The computing device searches the one or more seed questions against all questions in the Q&A library and identifies questions related to the topic. The computing device clusters the text of the questions related to the topic into a plurality of clusters and then removes substantial duplicates from the plurality of clusters. The computing device generates a rich content webpage by aggregating a question from each cluster onto a single webpage containing the topic.

Abstract: A computer-implemented method of managing questions and answers on a computer-hosted service. The method includes a computing device receiving text based tax question and answer pairings and inputting the tax question and answer pairings into a content model executed by the device and outputting a content score for each tax question and answer pairing based on the model. The content score comprises a number within a range. One end of the range corresponds to product content and another end of the range corresponds to general tax content. The device outputs an answer quality score for the tax question and answer pairings based at least in part on the content score and votes assigned to each respective question and answer pair, wherein votes comprises up votes and down votes. The device may generate a FAQ list stored in a database based at least in part on the answer quality score.

Abstract: A method for comparing purchase patterns includes matching multiple products purchased by a base company to multiple leaf nodes in a taxonomy tree to obtain multiple matching leaf nodes. The taxonomy tree is a hierarchical classification of products. The method further includes assigning, to each of the matching leaf nodes and to each parent node of the matching leaf nodes, a point value to obtain multiple point values, creating, for the base company and by a computer processor, a base feature vector including the point values, and calculating, by the computer processor, a similarity score between the base feature vector of the base company to a test feature vector of a test company. The method further includes providing, in response to the similarity score satisfying a similarity threshold, a recommendation.

Abstract: A method and system for selecting a product to advertise. The method includes receiving an advertisement request from an application, generating a plurality of nodes corresponding to a plurality of user-entered text strings received from a user by the application, sending, to a marketplace system, a plurality of search queries for the plurality of user-entered text strings, and receiving a plurality of product identifier in response to the plurality of search queries. The method further includes determining a plurality of edges corresponding to the plurality of product identifiers, generating a cluster using the plurality of nodes and the plurality of edges, selecting a product identifier of the plurality of product identifiers to obtain a selected product identifier, and providing, to the application, the selected product identifier, wherein the application displays, to the user, an advertisement for the product identified by the product identifier.

Abstract: A method and system for selecting a product to advertise. The method includes receiving an advertisement request from an application, generating a plurality of nodes corresponding to a plurality of user-entered text strings received from a user by the application, sending, to a marketplace system, a plurality of search queries for the plurality of user-entered text strings, and receiving a plurality of product identifier in response to the plurality of search queries. The method further includes determining a plurality of edges corresponding to the plurality of product identifiers, generating a cluster using the plurality of nodes and the plurality of edges, selecting a product identifier of the plurality of product identifiers to obtain a selected product identifier, and providing, to the application, the selected product identifier, wherein the application displays, to the user, an advertisement for the product identified by the product identifier.

Abstract: A unified framework to understand multimedia signals utilizes the loosely annotated multimedia data on the Web, analysis it in various signal domains, such as text, image, audio and combinations thereof, and builds an association graph called the “Multimedia Brain,” which basically comprises visual signals, audio signals, text phrases and the like that capture a multitude of objects, experiences and their attributes and the links among them that capture similar intent or functional and contextual relationships.

Abstract: The present disclosure describes a method and system called “Universal Learner (UL),” which provides a unified framework to understand multimedia signals. The UL utilizes the loosely annotated multimedia data on the Web, analyses it in various signal domains, such as text, image, audio and combinations thereof, and builds an association graph called the “Multimedia Brain,” which basically comprises visual signals, audio signals, text phrases and the like that capture a multitude of objects, experiences and their attributes and the links among them that capture similar intent or functional and contextual relationships.

Abstract: “A Universal Learner (UL)” provides a unified framework to understand multimedia signals. The UL utilizes the loosely annotated multimedia data on the Web, analyses it in various signal domains, such as text, image, audio and combinations thereof, and builds an association graph called the “Multimedia Brain,” which basically comprises visual signals, audio signals, text phrases and the like that capture a multitude of objects, experiences and their attributes and the links among them that capture similar intent or functional and contextual relationships.

Abstract: Methods and systems for discovering styles via color and pattern co-occurrence are disclosed. According to one embodiment, a computer-implemented method comprises collecting a set of fashion images, selecting at least one subset within the set of fashion images, the subset comprising at least one image containing a fashion item, and computing a set of segments by segmenting the at least one image into at least one dress segment. Color and pattern representations of the set of segments are computed by using a color analysis method and a pattern analysis method respectively. A graph is created wherein each graph node corresponds to one of a color representation or a pattern representation computed for the set of segments. Weights of edges between nodes of the graph indicate a degree of how the corresponding colors or patterns complement each other in a fashion sense.

Abstract: The present disclosure describes a method and system called “Universal Learner (UL),” which provides a unified framework to understand multimedia signals. The UL utilizes the loosely annotated multimedia data on the Web, analyses it in various signal domains, such as text, image, audio and combinations thereof, and builds an association graph called the “Multimedia Brain,” which basically comprises visual signals, audio signals, text phrases and the like that capture a multitude of objects, experiences and their attributes and the links among them that capture similar intent or functional and contextual relationships.