Aesthetics-guided image enhancement
Methods and systems are provided for generating enhanced image. A neural network system is trained where the training includes training a first neural network that generates enhanced images conditioned on content of an image undergoing enhancement and training a second neural network that designates realism of the enhanced images generated by the first neural network. The neural network system is trained by determine loss and accordingly adjusting the appropriate neural network(s). The trained neural network system is used to generate an enhanced aesthetic image from a selected image where the output enhanced aesthetic image has increased aesthetics when compared to the selected image.
1. One or more non-transitory computer-readable media having a plurality of executable instructions embodied thereon, which, when executed by one or more processors, cause the one or more processors to perform a method, the method comprising:
obtaining scored images, wherein a plurality of images are scored based on aesthetic attributes;
designating the scored images that have aesthetic scores within a predefined range as input images;
designating the scored images that have aesthetic scores above a predefined threshold as reference images;
training an aesthetic image enhancing neural network system using the input images with a first neural network and reference images with a second neural network, wherein the trained aesthetic image enhancing neural network system is used to generate an enhanced image from an image input;
analyzing portions of the scored images using a segmentation network, wherein analyzing portions of the scored images using the segmentation network causes adaptive adjustments by the first neural network; and
outputting the enhanced image, wherein the enhanced image has increased aesthetics when compared with the image input into the trained aesthetic image enhancing neural network system.
2. The non-transitory media of claim 1 ,
wherein the adaptive adjustments by the first neural network are dependent on the content of the scored images.
3. The non-transitory media of claim 1 , wherein the aesthetic scores are assigned based on one or more aesthetic attribute.
4. The non-transitory media of claim 3 , wherein the one or more aesthetic attributes include one or more of interesting content, object emphasis, lighting, color harmony, vivid color, shallow depth of field, motion blur, rule of thirds, balancing element, repetition, and symmetry.
5. The non-transitory media of claim 1 , wherein training the aesthetic image enhancing neural network system further comprises:
during a first iteration, determining loss using a comparison between a training enhanced image and an input image based in part on differences conditioned on a corresponding segmentation map; and
adjusting the aesthetic image enhancing neural network system based on the determined loss.
6. The non-transitory media of claim 5 , wherein the loss is one or more of content loss and adversarial loss.
7. The non-transitory media of claim 1 , wherein training the aesthetic image enhancing neural network system further comprises:
during a first iteration, determining loss using a comparison between a reference image and realism designation; and
adjusting the aesthetic image enhancing neural network system based on the determined loss.
8. The non-transitory media of claim 7 , wherein the loss is one or more of content loss and adversarial loss.
9. A computing system comprising:
one or more processors; and
one or more non-transitory computer-readable storage media, coupled with the one or more processors, having instructions stored thereon, which, when executed by the one or more processors, cause the computing system to:
obtain a set of images for training a first neural network of an aesthetic image enhancing neural network system;
determine a plurality of aesthetic scores for the set of images;
analyze portions of the scored images using a segmentation network, thereby causing adaptive adjustments by the first neural network;
designate a subset of the set of images, having aesthetic scores above a threshold level, as reference images for training a second neural network of the aesthetic image enhancing neural network system; and
train the aesthetic image enhancing neural network system using the set of images and the reference images, wherein the trained aesthetic image enhancing neural network system generates an enhanced image from a first input image.
10. The computing system of claim 9 , further comprising:
obtain a parsed image corresponding to the first input image; and
input the parsed image into the trained aesthetic image enhancing neural network system, wherein the first neural network converts the first input image into the enhanced aesthetic image using the parsed image.
11. The computing system of claim 9 , wherein the first neural network acts as a generator in the trained aesthetic image enhancing neural network system based on a generative adversarial type architecture.
12. The computing system of claim 9 , wherein the trained aesthetic image enhancing neural network system further includes a deactivated second neural network that acts as a discriminator during training.
13. The computing system of claim 9 , wherein the first input image is selected from a set of images stored on a user device, the set of images taken using a camera function of the user device.
14. The computing system of claim 9 , wherein the first neural network parses the set of images using a pyramid parsing network.
15. The computing system of claim 9 , wherein the first neural network parses the set of images using grayscale.