Back to results
US 10,776,626 B1Granted
Machine learning based identification of visually complementary item collections
AssigneeAmazon Technologies, Inc.
InventorsQiuying Lin, Brian Mann
Life of this patent
Estimating the term…
Abstract
Aspects of the present disclosure relate to machine learning techniques for identifying collections of items that are visually complementary. These techniques can relying on computer vision and item imagery. For example, a first portion of a machine learning system can be trained to extract aesthetic item qualities from pixel values of images of the items. A second portion of the machine learning system can learn correlations between these extracted aesthetic qualities and the level of visual coordination between items. Thus, the disclosed techniques use computer vision machine learning to programmatically determine whether items visually coordinate with one another based on pixel values of images of those items.
Claims24
A computing system comprising:
- at least one electronic data repository storing:
- item data including, for each item of a plurality of items of an electronic catalog of items, an image of the item and a category of a plurality of categories associated the item;
- and
- parameters of a machine learning system including (1) an attribute extraction model trained to determine aesthetic attributes of an item based on an analysis of an image of the item, wherein the aesthetic attributes include a line type attribute that is not defined in the item data, wherein the attribute extraction model is trained to assign to the line type attribute a value that specifies a type of line detected in the image, and (2) an artificially intelligent stylist model trained to identify particular ones of the plurality of items for inclusion in a visually complementary collection;
- a computer-readable memory storing instructions for generating the visually complementary collection;
- and
- one or more processors in communication with the at least one electronic data repository and the computer-readable memory, the one or more processors configured by the instructions to at least:
- identify a source item of the plurality of items to use as a basis for generating the visually complementary collection, the source item depicted by a first image, wherein the source item is associated with a first category of the plurality of categories;
- assign a first value of the line type attribute to the source item based on analyzing pixel values of the first image using the attribute extraction model;
- for each item of a set of items of different categories than the first category:
- assign an additional value of the line type attribute to the item based on analyzing pixel values of an image depicting the item, and
- determine a likelihood that the item matches the source item at least partly by providing the first value and the additional value as input into the artificially intelligent stylist model;
- generate a user interface for presenting the visually complementary collection, the user interface including the first image of the source item and a plurality of windows for depicting items of the different categories;
- populate the plurality of windows with images of particular ones of the set of items, the particular ones of the set of items identified based on the determined likelihoods;
- and
- cause output the user interface for presentation to a user.
13 more elements
A computer-implemented method, comprising:
- identifying a source item to use as a basis for generating a visually complementary collection of items, the source item depicted by a first image and associated with a first category a plurality of categories;
- assigning a first value of a line type attribute to the first item based on analyzing the first image using a trained attribute extraction model, the line type attribute representing a type of line detected;
- identifying a second category of the plurality of categories to be included in the visually complementary collection;
- identifying a second item associated with the second category, the second item depicted by a second image;
- assigning a second value of the line type attribute to the second item based on analyzing the second image using the trained attribute extraction model;
- providing the first and second values of the line type attribute to an artificially intelligent stylist model trained to identify visually complementary items;
- and
- based on an output of the artificially intelligent stylist model indicating that the first and second items are predicted to be visually complementary, generating a user interface presenting the first image of the source item and the second image of the second item as at least a portion of the visually complementary collection.
5 more elements
A non-transitory computer-readable medium storing instructions that, when executed, cause at least one processor to perform operations comprising:
- identifying a source item to use as a basis for generating a visually complementary collection of items, the source item depicted by a first image and associated with a first category a plurality of categories, wherein the first item is assigned a first value of a line type attribute representing a type of line detected, wherein the first value is determined from the first image using a trained attribute extraction model;
- identifying a second category of the plurality of categories to be included in the visually complementary collection;
- identifying a second item associated with the second category, the second item depicted by a second image, wherein the second item is assigned a second value of the line type attribute, wherein the second value is determined from the second image using the trained attribute extraction model;
- providing the first and second values of the line type attribute to an artificially intelligent stylist model trained to identify visually complementary items;
- and
- based on an output of the artificially intelligent stylist model indicating that the first and second items are predicted to be visually complementary, generating a user interface presenting the first image of the source item and the second image of the second item as at least a portion of the visually complementary collection.
3 more elements
Dependencies are inferred from the claim text.