Multimodal color variations using learned color distributions
Embodiments are disclosed for generating multiple color theme variations from an input image using learned color distributions. A method of generating multiple color theme variations from an input image using learned color distributions includes obtaining, by a user interface manager, an input image, determining, by a color extraction manager, one or more color priors based on the input image, generating, by a color distribution modeling network, a plurality of color theme variations based on the one or more color priors, ranking, by a color theme evaluation network, the plurality of color theme variations, and generating, by a recolor manager, a plurality of recolored output images using the plurality of color theme variations.
1. A computer-implemented method comprising:
obtaining, by a user interface manager, an input image;
determining, by a color extraction manager, one or more color priors based on the input image;
encoding the one or more color priors into an input tensor that represents color space values and weight values associated with each of the one or more color priors;
predicting, by a color distribution modeling network, a color space value of a color of a color theme variation, wherein the color is based on the one or more color priors, wherein the color distribution modeling network has been trained to model a color distribution of a training dataset, the training dataset including a plurality of color images having different color themes;
generating, by the color distribution modeling network, a plurality of color theme variations by iteratively predicting one or more colors of each color theme variation of the plurality of color theme variations;
ranking, by a color theme evaluation network, the plurality of color theme variations to obtain a plurality of ranked color theme variations; and
recoloring, by a recolor manager, the input image to generate a plurality of recolored output images using a number of top ranked color theme variations of the plurality of ranked color theme variations, wherein recoloring includes changing pixel color values of the input image according to the plurality of color theme variations.
2. The computer-implemented method of claim 1 , wherein determining, by a color extraction manager, one or more color priors based on the input image, further comprises:
identifying a plurality of unique colors in the input image;
clustering the plurality of unique colors into a plurality of clusters; and
determining a color theme associated with the input image, wherein each color of the color theme is associated with each of the plurality of clusters.
3. The computer-implemented method of claim 2 , further comprising:
sampling a subset of colors from the color theme to use as the one or more color priors.
4. The computer-implemented method of claim 1 , wherein the color space value includes a next color space value or weight value.
5. The computer-implemented method of claim 4 , wherein the color space value includes the next color space value, further comprising:
updating the input tensor to include the next color space value.
6. The computer-implemented method of claim 1 , wherein ranking, by a color theme evaluation network, the plurality of color theme variations, further comprises:
receiving the plurality of color theme variations;
predicting a score for each color theme variation; and
ranking the plurality of color theme variations based on each color theme's predicted score.
7. The computer-implemented method of claim 6 , wherein predicting a score for each color theme variation, further comprises:
predicting a plurality of scores for a plurality of subsets of colors of a first color theme variation; and
determining a score for the first color theme variation by combining the plurality of scores.
8. The computer-implemented method of claim 1 , wherein the color theme evaluation network is trained by a training manager to predict a likelihood that an input color theme was included in a training dataset and to generate a score based on the likelihood.
9. A system, comprising:
at least one processor; and
a memory including instructions stored thereon which, when executed by the at least one processor, cause the system to:
obtain an input image;
determine one or more color priors based on the input image;
encode the one or more color priors into an input tensor that represents color space values and weight values associated with each of the one or more color priors;
predict, using a first machine learning model, a color space value of a color of a color theme variation, wherein the color is based on the one or more color priors, and wherein the first machine learning model has been trained to model a color distribution of a training dataset, the training dataset including a plurality of color images having different color themes;
generate a plurality of color theme variations by iteratively predicting one or more colors of each color theme variation of the plurality of color theme variations;
rank, using a second machine learning model, the plurality of color theme variations to obtain a plurality of ranked color theme variations; and
recolor the input image to generate a plurality of recolored output images using a number of top ranked color theme variations of the plurality of ranked color theme variations, wherein recoloring includes changing pixel color values of the input image according to the plurality of color theme variations.
10. The system of claim 9 , wherein to determine one or more color priors based on the input image, the system is further configured to:
identify a plurality of unique colors in the input image;
cluster the plurality of unique colors into a plurality of clusters; and
determine a color theme associated with the input image, wherein each color of the color theme is associated with each of the plurality of clusters.
11. The system of claim 10 , wherein the system is further configured to sample a subset of colors from the color theme to use as the one or more color priors.
12. The system of claim 9 , wherein the color space value includes a next color space value or weight value.
13. The system of claim 12 , wherein the color space value includes the next color space value and wherein the system is further configured to:
update the input tensor to include the next color space value.
14. The system of claim 9 , wherein to rank the plurality of color theme variations, the system is further configured to:
receive the plurality of color theme variations;
predict a score for each color theme variation; and
rank the plurality of color theme variations based on each color theme's predicted score.
15. The system of claim 14 , wherein to predict a score for each color theme variation, the color theme evaluation network second machine learning model is further configured to:
predict a plurality of scores for a plurality of subsets of colors of a first color theme variation; and
determine a score for the first color theme variation by combining the plurality of scores.
16. The system of claim 9 , wherein the second machine learning model is trained to predict a likelihood that an input color theme was included in a training dataset and to generate a score based on the likelihood.
17. A system, comprising:
means for obtaining an input image;
means for determining one or more color priors based on the input image;
means for encoding the one or more color priors into an input tensor that represents color space values and weight values associated with each of the one or more color priors;
means for predicting a color space value of a color of a color theme variation using a machine learning model, wherein the color is based on the one or more color priors and wherein the machine learning model has been trained to model a color distribution of a training dataset, the training dataset including a plurality of color images having different color themes;
means for generating a plurality of color theme variations by iteratively predicting one or more colors of each color theme variation of the plurality of color theme variations;
means for ranking the plurality of color theme variations to obtain a plurality of ranked color theme variations; and
means for recoloring the input image to generate a plurality of recolored output images using a number of top ranked color theme variations of the plurality of ranked color theme variations.
18. The system of claim 17 , wherein the color space value includes a next color space value or weight value, wherein recoloring includes changing pixel color values of the input image according to the plurality of color theme variations.
19. The system of claim 18 , wherein the color space value includes the next color space value, further comprising:
means for updating the input tensor to include the next color space value.
20. The system of claim 17 , wherein the means for ranking the plurality of color theme variations, further comprises:
means for receiving the plurality of color theme variations;
means for predicting a score for each color theme variation; and
means for ranking the plurality of color theme variations based on each color theme's predicted score.