	US 12,462,081 B2
	Systems and methods for generating interior designs using artificial intelligence
Ankur Agrawal, Chicago, IL (US); Vinod Koppurapu, Fremont, CA (US); and Vinay Bhardwaj, Orinda, CA (US)
Assigned to Inhabitr, Inc.
Filed by Inhabitr, Inc., Chicago, IL (US)
Filed on Mar. 13, 2024, as Appl. No. 18/604,423.
Claims priority of provisional application 63/451,777, filed on Mar. 13, 2023.
Prior Publication US 2024/0311533 A1, Sep. 19, 2024
Int. Cl. G06F 7/48 (2006.01); G06F 30/12 (2020.01); G06F 30/13 (2020.01); G06F 30/27 (2020.01); G06F 111/16 (2020.01); G06F 111/18 (2020.01)

CPC G06F 30/27 (2020.01) [G06F 30/12 (2020.01); G06F 30/13 (2020.01); G06F 2111/16 (2020.01); G06F 2111/18 (2020.01)]

15 Claims

1. A method of generating interior design options, comprising:

obtaining an inspiration image of a portion of a room;

parsing the inspiration image using a plurality of trained machine learning models to:

(a) identify and segment one or more furniture, fixture or equipment items into sub-images, and (b) extract a plurality of attributes for each identified furniture, fixture or equipment item from the sub-images,

wherein each model of the plurality of trained machine learning models is trained on a distinct interior design segment within hospitality and commercial office sectors, the segments comprising hotels, student housing, multi-family residences, and commercial offices, wherein each model of the plurality of trained machine learning models is trained using a training dataset comprising images with furniture, fixture or equipment having variations in size, style, shape and texture, and rooms having variations in lighting, color, ambience and layout within its respective interior design segment, and wherein each model of the plurality of trained machine learning models is trained to analyze space types including living rooms, bedrooms and common areas;

identifying alternatives corresponding to each furniture, fixture or equipment by:

(a) querying a furniture data cloud storing inventory and product catalogs from a network of suppliers, manufacturers and distributors, using the extracted attributes,

(b) considering user-provided budget constraints and style preferences, and

generating and displaying variations of the room with images corresponding to the alternatives corresponding to each furniture, fixture or equipment, wherein generating the variations of the room comprises using a trained generative adversarial network, a trained autoencoder, and a trained diffusion model to generate images with different orientations and/or texture, based on features extracted from images corresponding to the alternatives;

creating separate classes for each furniture, feature and equipment category;

performing object extraction using bounding boxes for each of the furniture, feature and equipment categories;

performing background removal of cropped images;

using precision and recall as metrics for evaluating a performance of the plurality of trained machine learning models; and

fine tuning the plurality of trained machine learning models using image augmentation to increase diversity and applying random transformations in response to the evaluation not meeting a performance threshold whereby more diverse data results in a better training set and whereby a predictive model is continuously trained to make generative output related to the furniture, feature and/or equipment categories more accurate.