Generative Machine Learning For Fashion

ODOMOJULI

2019-12-18T08:00:00.000Z

Generative Machine Learning For Fashion

We propose an alternate method of architecture for generative adversarial networks. The architecture uses unsupervised separation of high-level attributes of clothing and stochastic variation in generated images. This enables intuitive, scale-specific control of image synthesis. This generator performs comfortably within traditional distribution quality metrics and has demonstrably decent interpolation properties. Latent factors of variation are distinct and accessible. Of course, the quality of the model depends on the dataset.

© 2020 ODOMOJULI