We are already aware of machine learning codes disrupting the fashion retail market such as inventory management, virtual reality systems for apparel fitting and most commonly, recommendation systems based on consumer preferences, buying behavior, etc. I myself captured some of these in the post on ML applications in the e-commerce ecosystem, written a few months ago. Today, on a separate note, we’ll look at some of the initiatives where machine learning is disrupting the consumer fashion market in some really novel and unique ways. In a way, these models are planning for a future where machines will be able to interpret art the way we do.
Researchers at University of Toronto are working at building a machine learning model to help you improve your “fashionability”. Fashionability implies not only what you wear but encompassing everything that forms part of the clicked image – including the individual in the image, the picture shot, the background, etc. The algorithm will evaluate your image on its overall visual appeal in comparison with other evaluated images in its database. So called ‘evaluated’ images refer to analysis derived from more than 144,000 likes and comments received on posts on the fashion website chictopia.com. For the new images, this group constitutes the training dataset. How it works? The model uses Orbeus ReKognition API to score facial elements, metadata scan of the outfit post, background analysis using trained scene classifiers, data from Flickr80k image styles and location data into the model feed to give a recommendation on “fashionability” of the image. In the future, the model should also be able to recommend what to wear in a particular setting.
Finding the Next Big “Supermodel” Face
Researchers haven’t limited their focus on inanimate objects. Researchers at the Indiana University at Bloomington are focusing on building machine learning algorithms to identify future supermodels. The models input data ranging from physical attributes of the fashion models, their agencies, their runway appearances and their social media activity such as Instagram. Correlating data, the models are predicting how factors such as extra inches of height, modelling agency, physical attributes, social media activity, etc. impact future successes of new faces in the fashion modelling industry. Interestingly, results so far indicate that physical attributes aren’t the top consideration factors for potential success.
Machine as the “Fashion Designer”
Machine learning intervention isn’t limited to merely predicting in the fashion industry. TJ Torres, founder at the CA based startup Stitch Fix, recently wrote a paper on how his firm is using the power of neural networks to create a future where machines will be the fashion designers. The test models are creating appealing machine generated print designs for the apparels, using fed data on existing designs. Elsewhere, there are published research on using genetic algorithms for building similar models.
With such initiatives on the anvil, machines taking over the genre of “art” also looks like a possibility for the future.