Personalized digital image aesthetics in a digital medium environment
Techniques and systems are described to determine personalized digital image aesthetics in a digital medium environment. In one example, a personalized offset is generated to adapt a generic model for digital image aesthetics. A generic model, once trained, is used to generate training aesthetics scores from a personal training data set that corresponds to an entity, e.g., a particular user, group of users, and so on. The image aesthetics system then generates residual scores (e.g., offsets) as a difference between the training aesthetics score and the personal aesthetics score for the personal training digital images. The image aesthetics system then employs machine learning to train a personalized model to predict the residual scores as a personalized offset using the residual scores and personal training digital images.
1. In a digital medium environment for personalized digital image aesthetics, a method implemented by at least one computing device, the method comprising:
generating, by the at least one computing device, a generic aesthetics score for a digital image using at least one generic model trained using machine learning on a generic training data set;
generating, by the at least one computing device, a personalized offset for the digital image using at least one personalized model trained using machine learning on a personal training data set associated with an entity;
determining, by the at least one computing device, a personalized aesthetics score for the digital image based on the personalized offset and the generic aesthetics score; and
outputting, by the at least one computing device, the personalized aesthetics score for the digital image.
2. The method as described in claim 1 , wherein: the at least one personalized model is trained using machine learning based on a deviation of a personal aesthetics score with respect to a training aesthetics score;
the personal aesthetics score is specified by the entity as part of the personal training data set; and
the training aesthetics score is generated by the at least one generic model from the personal training data set.
3. The method as described in claim 1 , wherein the generating of the personalized offset includes:
extracting content attributes from the digital image using at least one content model, of the at least one personalized model, trained using machine learning on the personal training data set;
extracting aesthetic attributes from the digital image using at least one aesthetic model, of the at least one personalized model, trained using machine learning on the personal training data set; and
generating the personalized offset for the entity based on the extracted content attributes and the extracted aesthetic features using machine learning.
4. The method as described in claim 3 , wherein the generating of the personalized offset includes applying a support vector machine on the extracted aesthetic attributes and the extracted content attributes.
5. The method as described in claim 1 , wherein the personal training data set is configured by automatically selecting representative personal training digital images from a plurality of personal training digital images using active learning, each of the plurality of personal training digital images are associated with a personal aesthetics score specified by the entity through interaction with a user interface.
6. The method as described in claim 5 , wherein the active learning includes selecting the representative personal training digital images automatically and without user intervention, by the at least one computing device, based on aesthetic style diversity and reduction in redundancy.
7. The method as described in claim 5 , wherein:
the active learning includes selecting the representative personal training digital images automatically and without user intervention by maximizing a deviation of the personal aesthetics scores with respect to training aesthetics scores; and
the training aesthetics scores are generated by the at least one generic model using machine learning based on the plurality of personal training digital images.
8. The method as described in claim 1 , further comprising generating a personalized image result based at least in part on the personalized aesthetics score for a plurality of said digital images.
9. The method as described in claim 8 , wherein the personalized image result is part of image curation or an image search result.
10. In a digital medium environment to support personalized digital image aesthetics training, a system comprising:
a generic aesthetics module implemented at least partially in hardware of a computing device to generate training aesthetics scores from a personal training data set using a generic model trained using machine learning on a generic training data set;
an offset determination module implemented at least partially in hardware of the computing device to determine residual scores based on a difference between the training aesthetics scores and respective user-specified personal aesthetics scores associated with the personal training set; and
a machine learning module implemented at least partially in hardware of the computing device to train at least one personalized model using machine learning to generate a personalized offset based at least in part on the determined residual scores.
11. The system as described in claim 10 , further comprising an active learning module to generate the personal training data set by:
automatically selecting representative training digital images from a plurality of training digital images; and
obtaining the user-specified personal aesthetics scores from the entity via a user interface for the selected representative training digital images.
12. The system as described in claim 11 , wherein the active learning module selects the representative training digital images automatically from the plurality of training digital images based on aesthetic style diversity and reduction in redundancy.
13. The system as described in claim 11 , wherein the active learning module selects the representative training digital images by maximizing a deviation of the personal aesthetics scores with respect to training aesthetics scores generated by the at least one generic model as applied to the plurality of training digital images.
14. The system as described in claim 10 , further comprising an image management module to generate a personalized image result based at least in part on a personalized aesthetics score generated from a personalized offset for a subsequent digital image by the at least one personalized model using machine learning.
15. The system as described in claim 10 , wherein the offset determination module is configured to generate of the personalized offset by:
extracting content attributes from the digital image using at least one content attribute model trained using machine learning on the personal training data set;
extracting aesthetic attributes from the digital image using at least one aesthetic attribute model content trained using machine learning on the personal training data set; and
generating the personalized offset using machine learning based on the extracted content attributes and the extracted aesthetic features.
16. In a digital medium environment for personalized digital image aesthetics, a system comprising:
means for generating a generic aesthetics score for a digital image using at least one generic model trained using machine learning on a generic training data set;
means for generating a personalized offset for the digital image using at least one personalized model trained using machine learning, the generating means including:
means for extracting content attributes from the digital image using at least one content attribute model, of the at least one personalized model, trained using machine learning on a personal training data set;
means for extracting aesthetic attributes from the digital image using at least one aesthetic attribute model, of the at least one personalized model, trained using machine learning on the personal training data set; and
means for generating the personalized offset for the entity based on the content attributes and the aesthetic features using machine learning; and
means for determining a personalized aesthetics score for the digital image based on the personalized offset and the generic aesthetics score.
17. The system as described in claim 16 , wherein:
the at least one personalized model is trained using machine learning based on a deviation of a personal aesthetics score with respect to a training aesthetics score;
the personal aesthetics score is specified by the entity as part of the personal training data set; and
the training aesthetics score is generated by the at least one generic model from the personal training data set.
18. The system as described in claim 16 further comprising means for active learning to automatically select representative personal training digital images from a plurality of personal training digital images, each of the plurality of personal training digital images are associated with a personal aesthetics score specified by the entity through interaction with a user interface.
19. The system as described in claim 18 , wherein the active learning means includes selecting the representative personal training digital images automatically and without user intervention based on aesthetic style diversity, reduction in redundancy, or to maximize a deviation of the personal aesthetics scores with respect to training aesthetics scores, the training aesthetics scores generated by the at least one generic model using machine learning based on the plurality of personal training digital images.
20. The system as described in claim 16 , wherein the generating means includes means for applying a support vector machine on the extracted aesthetic attributes and the extracted content attributes.