US 12,475,666 B2
Systems and methods for defurnishing and furnishing spaces, and removing objects from spaces
David Alan Gausebeck, Sunnyvale, CA (US); Japjit Tulsi, Sunnyvale, CA (US); Rj Pittman, Sunnyvale, CA (US); David Lippman, Sunnyvale, CA (US); Vivek Tanna, Sunnyvale, CA (US); Nicole Guernsey, Sunnyvale, CA (US); and Olaf Brandt, Sunnyvale, CA (US)
Assigned to CoStar Realty Information, Inc., Arlington, VA (US)
Filed by CoStar Realty Information, Inc., Arlington, VA (US)
Filed on Jun. 13, 2024, as Appl. No. 18/742,947.
Claims priority of provisional application 63/567,185, filed on Mar. 19, 2024.
Claims priority of provisional application 63/472,795, filed on Jun. 13, 2023.
Prior Publication US 2024/0420437 A1, Dec. 19, 2024
Int. Cl. G06T 19/20 (2011.01); G06T 17/20 (2006.01); G06V 10/82 (2022.01); G06V 20/70 (2022.01)
CPC G06T 19/20 (2013.01) [G06T 17/20 (2013.01); G06V 10/82 (2022.01); G06V 20/70 (2022.01); G06T 2200/24 (2013.01); G06T 2210/04 (2013.01); G06T 2219/004 (2013.01); G06T 2219/2024 (2013.01)] 26 Claims
OG exemplary drawing
 
1. A non-transitory computer-readable medium comprising executable instructions, the executable instructions being executable by one or more processors to perform a method, the method comprising:
accessing data of a multidimensional space representing a physical environment, the data including input 2D images of the physical environment;
identifying interior elements within the input 2D images using a machine learning model, at least one of the interior elements representing furniture in the physical environment;
masking one or more of the interior elements within the input 2D images with masks;
filling each of the masks with imagery of the physical environment to generate filled 2D images, wherein filling includes utilizing a latent diffusion model to fill each of the masks with imagery of the physical environment to generate the filled 2D images, wherein the latent diffusion model was trained using first 2D images of unfurnished spaces, and wherein the latent diffusion model was fine-tuned using second 2D images of unfurnished spaces modified to include artifacts not present in the second 2D images of unfurnished spaces;
generating, based on the input 2D images and the filled 2D images, output 2D images that create an appearance of a defurnished space, the defurnished space having the one or more interior elements appearing as missing from the multidimensional space representing the physical environment;
providing, using the output 2D images, all or some of the defurnished space for display;
generating one or more additional interior elements within the defurnished space based at least on characteristics of the one or more additional interior elements;
positioning the one or more additional interior elements within the defurnished space based on a geometry of the defurnished space and a layout rule, the layout rule being applied to a position or orientation of the one or more additional interior elements within the defurnished space; and
providing all or some of the defurnished space with the one or more additional interior elements positioned within the defurnished space for display.