Systems and methods for defurnishing and furnishing spaces, and removing objects from spaces
An example system may access data of a multidimensional space representing a physical environment and identify interior elements within the multidimensional space using a first machine learning model. The interior element may represent furniture in the physical environment. The system may mask one or more of the interior elements with masks and fill each of the masks with imagery of the physical environment to 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. The system may provide all or some of the defurnished space for display.
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.
2 . The non-transitory computer-readable medium of claim 1 , wherein the physical environment is a furnished room.
3 . The non-transitory computer-readable medium of claim 1 , wherein the multidimensional space is a 2D representation of the physical environment.
4 . The non-transitory computer-readable medium of claim 3 , wherein the 2D representation of the physical environment is used to generate a corresponding 3D representation of the physical environment.
5 . The non-transitory computer-readable medium of claim 1 , wherein the multidimensional space is a 3D representation of the physical environment.
6 . The non-transitory computer-readable medium of claim 1 , wherein the data of the multidimensional space further includes a textured 3D mesh.
7 . The non-transitory computer-readable medium of claim 1 , wherein the interior elements within the multidimensional space further include at least one wall within the physical environment.
8 . The non-transitory computer-readable medium of claim 1 , wherein the machine learning model includes a semantic segmentation neural network.
9 . The non-transitory computer-readable medium of claim 1 , wherein filling each of the masks with the imagery of the physical environment comprising applying inpainting to the masks.
10 . The non-transitory computer-readable medium of claim 1 , the method further comprising:
receiving a selection of at least one design style type from a user; and
selecting interior elements within the multidimensional space that were previously identified based on the at least one design style type from the user, wherein masking the one or more interior elements comprises masking the one or more interior elements that are of the at least one design style type.
11 . The non-transitory computer-readable medium of claim 1 , the method further comprising:
receiving a selection of at least one design style type from a user; and
selecting interior elements within the multidimensional space that were previously identified based on the at least one design style type from the user, wherein masking the one or more interior elements comprises masking the one or more interior elements that are not of the at least one design style type.
12 . The non-transitory computer-readable medium of claim 1 , wherein the one or more additional interior elements to be added to the defurnished space are requested by a user.
13 . The non-transitory computer-readable medium of claim 12 , the method further comprising:
receiving an additions request from the user, comprising receiving a prompt from a user; and
applying the prompt to a large language model to receive a response from the large language model, the response identifying the one or more additional interior elements.
14 . A system comprising at least one processor and memory containing executable instructions, the executable instructions being executable by the at least one processor to:
access data of a multidimensional space representing a physical environment, the data including input 2D images of the physical environment;
identify 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;
mask at least some of the interior elements within the input 2D images with masks;
fill 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 at least some of the interior elements appearing as missing from the multidimensional space representing the physical environment;
provide, using the output 2D images, all or some of the defurnished space for display;
generate one or more additional interior elements within the defurnished space based at least on characteristics of the one or more additional interior elements;
position 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
provide all or some of the defurnished space with the one or more additional interior elements positioned within the defurnished space for display.
15 . The system of claim 14 , wherein the physical environment is a furnished room.
16 . The system of claim 14 , wherein the multidimensional space is a 2D representation of the physical environment.
17 . The system of claim 16 , wherein the 2D representation of the physical environment is used to generate a corresponding 3D representation of the physical environment.
18 . The system of claim 14 , wherein the multidimensional space is a 3D representation of the physical environment.
19 . The system of claim 14 , wherein the data of the multidimensional space further includes a textured 3D mesh.
20 . The system of claim 14 , wherein the interior elements within the multidimensional space further include at least one wall within the physical environment.
21 . The system of claim 14 , wherein the machine learning model includes a semantic segmentation neural network.
22 . The system of claim 14 , wherein filling each of the masks with the imagery of the physical environment comprising applying inpainting to the masks.
23 . The system of claim 14 , the executable instructions being further executable by the at least one processor to:
receive a selection of at least one design style type from a user; and
select interior elements within the multidimensional space that were previously identified based on the at least one design style type from the user, wherein masking at least some of the interior elements comprises masking at least some of the interior elements that are of the at least one design style type.
24 . The system of claim 14 , the executable instructions being further executable by the at least one processor to:
receive a selection of at least one design style type from a user; and
select interior elements within the multidimensional space that were previously identified based on the at least one design style type from the user, wherein masking at least some of the interior elements comprises masking at least some of the interior elements that are not of the at least one design style type.
25 . The system of claim 14 , wherein the one or more additional interior elements to be added to the defurnished space are requested by a user.
26 . The system of claim 25 , wherein the executable instructions being further configured by the at least one processor to:
receive an additions request from the user, comprising to receive a prompt from a user; and
apply the prompt to a large language model to receive a response from the large language model, the response identifying the one or more additional interior elements.