Transferring hairstyles between portrait images utilizing deep latent representations
The disclosure describes one or more embodiments of systems, methods, and non-transitory computer-readable media that generate a transferred hairstyle image that depicts a person from a source image having a hairstyle from a target image. For example, the disclosed systems utilize a face-generative neural network to project the source and target images into latent vectors. In addition, in some embodiments, the disclosed systems quantify (or identify) activation values that control hair features for the projected latent vectors of the target and source image. Furthermore, in some instances, the disclosed systems selectively combine (e.g., via splicing) the projected latent vectors of the target and source image to generate a hairstyle-transfer latent vector by using the quantified activation values. Then, in one or more embodiments, the disclosed systems generate a transferred hairstyle image that depicts the person from the source image having the hairstyle from the target image by synthesizing the hairstyle-transfer latent vector.
1 . A non-transitory computer-readable medium storing instructions that, when executed by at least one processor, cause the at least one processor to perform operations comprising:
receiving, via a graphical user interface, a user selection to transfer a hairstyle depicted on a first person in a first digital image to a second person depicted in a second digital image;
extracting a first set of latent features from the first digital image;
determining a subset of latent features from the first set of latent features extracted from the first digital image that corresponds to hair of the hairstyle depicted on the first person in the first digital image;
extracting a second set of latent features from the second digital image; and
generating, utilizing a generative neural network, a modified second digital image depicting the second person with the hairstyle of the first person from the first digital image by decoding a combination of the subset of latent features corresponding to hair of the hairstyle in the first digital image and the second set of latent features from the second digital image.
2 . The non-transitory computer-readable medium of claim 1 , wherein the operations further comprise receiving, via the graphical user interface, a user selection of the first digital image from an image upload or an image file selection.
3 . The non-transitory computer-readable medium of claim 1 , wherein the operations further comprise:
providing, for display within the graphical user interface, the first digital image, the second digital image, and a selectable option to initiate a hairstyle transfer; and
receiving, via the graphical user interface, the user selection to transfer the hairstyle depicted on the first person in the first digital image to the second person depicted in the second digital image by receiving a user interaction with the selectable option.
4 . The non-transitory computer-readable medium of claim 3 , wherein the operations further comprise providing, for display within the graphical user interface, the modified second digital image depicting the second person with the hair of the first person from the first digital image while displaying the first digital image and the second digital image.
5 . The non-transitory computer-readable medium of claim 1 , wherein the operations further comprise generating the modified second digital image depicting the second person with the hair of the first person from the first digital image by reconstructing facial features for one or more exposed facial regions of the second person when the hair of the hairstyle depicted on the first person in the first digital image is shorter than hair of the second person depicted in the second digital image.
6 . The non-transitory computer-readable medium of claim 1 , wherein the operations further comprise generating a blended modified second digital image by blending regions of the second digital image depicting non-hair facial features with the modified second digital image depicting the second person with the hair of the first person from the first digital image.
7 . The non-transitory computer-readable medium of claim 1 , wherein the operations further comprise:
generating a combined set of latent features by replacing one or more latent features from the second set of latent features with the first set of latent features; and
generating the modified second digital image by utilizing the generative neural network from the combined set of latent features.
8 . The non-transitory computer-readable medium of claim 1 , wherein the operations further comprise extracting the first set of latent features from the first digital image that correspond to hair of the hairstyle depicted on the first person in the first digital image utilizing noise pixels.
9 . A system comprising:
a memory component; and
a processing device coupled to the memory component, the processing device to perform operations comprising:
receiving, via a graphical user interface, a user selection of a first digital image depicting a first person with a hairstyle and a second digital image depicting a second person;
receiving, via the graphical user interface, a user interaction with a selectable option to transfer the hairstyle depicted on the first person in the first digital image to the second person depicted in the second digital image;
extracting a first set of latent features from the first digital image;
determining a subset of latent features from the first set of latent features extracted from the first digital image that corresponds to hair of the hairstyle;
extracting a second set of latent features from the second digital image;
generating, utilizing a generative neural network, a modified second digital image depicting the second person with the hairstyle of the first person from the first digital image by decoding a combination of the subset of latent features corresponding to the hair of the hairstyle in the first digital image and the second set of latent features from the second digital image; and
providing, for display within the graphical user interface, the modified second digital image.
10 . The system of claim 9 , wherein the operations further comprise providing, for display within the graphical user interface, the modified second digital image, the first digital image, and the second digital image.
11 . The system of claim 9 , wherein the operations further comprise reconstructing facial features for one or more exposed facial regions of the second person depicted in the second digital image within the modified second digital image based on the hair of the hairstyle depicted on the first person in the first digital image being a shorter length than hair of the second person depicted in the second digital image.
12 . The system of claim 9 , wherein the operations further comprise:
generating a set of transfer latent features by replacing one or more features from the second set of latent features from the second digital image with the subset of latent features from the first set of latent features extracted from the first digital image that correspond to hair of the hairstyle depicted on the first person in the first digital image; and
generating a synthesized digital image depicting the second person with the hair of the first person from the first digital image by utilizing the generative neural network with the set of transfer latent features.
13 . The system of claim 12 , wherein the operations further comprise generating the modified second digital image by blending the synthesized digital image with the second digital image, wherein blending comprises feathering, contrast blending, color blending, brightness blending, or additive blending.
14 . The system of claim 9 , wherein the operations further comprise determining the subset of latent features controlling depictions of hair in the first digital image by utilizing noise pixels in place of the hairstyle depicted in the first digital image.
15 . A computer-implemented method comprising:
receiving, via a graphical user interface, a user selection to transfer a hairstyle depicted on a first person in a first digital image to a second person depicted in a second digital image;
extracting a first set of latent features from the first digital image;
determining a subset of latent features from the first set of latent features extracted from the first digital image that corresponds to hair of the hairstyle depicted on the first person in the first digital image;
extracting a second set of latent features from the second digital image; and
generating, utilizing a generative neural network, a modified second digital image depicting the second person with the hairstyle of the first person from the first digital image by decoding a combination of the subset of latent features corresponding to the hair of the hairstyle in the first digital image and the second set of latent features from the second digital image.
16 . The computer-implemented method of claim 15 , further comprising providing, for display within the graphical user interface, the modified second digital image depicting the second person with the hair of the first person from the first digital image.
17 . The computer-implemented method of claim 15 , further comprising generating, utilizing the generative neural network, the modified second digital image depicting the second person with one or more additional facial attributes of the first person from the first digital image.
18 . The computer-implemented method of claim 15 , further comprising receiving, via the graphical user interface, a user selection of the first digital image and the second digital image from an image upload or an image file selection.
19 . The computer-implemented method of claim 15 , wherein extracting the subset of latent features from the first set of latent features extracted from the first digital image that corresponds to hair of the hairstyle depicted on the first person in the first digital image comprises utilizing a latent vector from the first digital image and an additional latent vector based on the first digital image with noise pixels in place of one or more regions depicting the hairstyle in the first digital image.
20 . The computer-implemented method of claim 15 , further comprising:
receiving, via the graphical user interface, an additional user selection to transfer an additional hairstyle depicted on a third person in a third digital image to the second person depicted in the second digital image; and
generating, utilizing the generative neural network from a third set of latent features and the second set of latent features, an additional modified second digital image depicting the second person with the additional hairstyle of the third person from the third digital image.