GANArtworks In the Mood

Auteur-es

YuLing Chen, Colorado Reed, Hongsuk Nam, Kevin Jun, Chenlin Ye, Joyce Shen, David Steier

Nombre Auteurs

7

Titre

GANArtworks In the Mood

Année de publication

2021

Référence (APA)

35th Conference on Neural Information Processing Systems (NeurIPS 2021), Sydney, Australia.

résumé

In home decoration, an artwork with a particular combination of colors can convey a positive mood that improves psychological health. In our work, we leverage emotion to color mapping techniques and Generative Adversarial Networks (GANs) to generate artwork that brings a room into a more positive mood. We create a unique workflow to extract the color scheme from a room photo, convert it to an image with target colors that represent the desired positive mood, and then use a conditional GAN to generate artwork with fine control of the color. In this paper, we share our emotion to color mapping pipeline, GAN model training, and evaluation results on the generated artworks.

URL

https://research.facebook.com/file/684243722573832/GANArtworks-In-the-Mood.pdf

Accessibilité de l'article

Libre

Champ

Artificial Intelligence, Machine Learning

Type contenu (théorique Applicative méthodologique)

Applicatif

Méthode

The method involves using emotion to color mapping techniques and GANs to generate artwork that brings a room into a more positive mood.

"We provide a web application [12] where a user uploads a room photo. Then we extract a color scheme from the photo, and then leverage an emotion-to-color map to determine an emotion. This map is a digitized version of Koboyashi’s studies [4,5,6,7], where a set of grouped emotions maps to a simplified set of 10 color hues and 12 shades within each hue. [...] We then use a weighted Euclidean distance function: l2_distance ∗ pixel_f raction to obtain the closest emotion to the user’s room. After the emotion of the user’s current room is detected, we provide a list of positive emotion groups for the user to select. Then we choose one emotion from the user selected group and impose the original room photo with the corresponding colors. Finally, we pass this imposed user room image into a conditional GAN to generate an artwork that conveys the desired positive mood."

Cas d'usage

N/A

Objectifs de l'article

The objectives of the article are to present a unique workflow for generating artwork that evokes positive feelings and improves home decoration.

Question(s) de recherche/Hypothèses/conclusion

The research question is how can we use emotion to color mapping techniques and GANs to generate artwork that brings a room into a more positive mood.

The hypothesis is that using emotion to color mapping techniques and GANs can generate artwork that evokes positive feelings and improves home decoration.

The conclusions are that the presented workflow is effective in generating artwork that brings a room into a more positive mood and improves home decoration.

Cadre théorique/Auteur.es

The theoretical framework of the article includes emotion to color mapping techniques and GANs. The main authors cited include the NIMH Center for the Study of Emotion and Attention, the Nippon Color & Design Research Institute, and WikiArt.

Concepts clés

Color mapping

Données collectées (type source)

"WikiArt impressionist landscape data (https://www.wikiart.org/) and landscape photo datasets downloaded from HistoGAN github (https://github.com/mahmoudnafifi/HistoGAN)."

Définition des émotions

Non

Ampleur expérimentation (volume de comptes)

10,000 impressionist paintings downloaded from the WikiArt website .

5,000 impressionist landscape paintings from the WikiArt website

5,000 landscape photos from the HistoGAN github repository

Technologies associées

GANs and emotion to color mapping techniques

Mention de l'éthique

Yes, there is mention of ethics in the source.
"Beyond the well known ethical implications of using any "deepfake" technologies, this particular project has no extended negative considerations. Creating artworks that promotes the moods of user’s rooms helps to improve people’s psychological health. Possibly, using the techniques that we described in this paper would make home decoration easier and more fun."

The authors state that they have read the ethics review guidelines and ensured that their paper conforms to them. They also discussed potential negative societal impacts of their work.

Both datasets are "open dataset with no personally identifiable information or offensive content."

Finalité communicationnelle

"Our experience in the image generation based on user selected positive moods shows huge potential in using generative modeling for artistic home decoration. As next steps, we shall continue improving mood mapping accuracy, exploring other artwork styles, and generating images with higher resolution."

Sell ultra personalized home decoration supposed to put people in better moods

Commentaires

Je ris