With the rise of the digital shopping age, second-hand retail websites are becoming increasingly popular, particularly within the fashion industry. Websites such as these allow users to upload listings of articles they hope to sell, often including images of the object for sale.
Photos taken by inexperienced photographers using unideal equipment such as a smartphone camera often have a very low aesthetic quality, an image feature that fashion websites cannot directly measure and prevent. In this work, we use human binary classifications of image aesthetic quality
to calculate popularity scores, which are then used to train an aesthetic quality predictor. Image features that correlate with aesthetic quality are extracted and utilized in a machine learning algorithm. With a regression output predicting a popularity score on a scale of 0 to 1 our method
proves to be a concise yet effective approach to predicting the aesthetic quality of fashion images. Our models proved effective and promising for future research. Our base model, trained with our entire dataset, resulted in an error value of only 18% in the most successful application. With
the ability to predict the aesthetic quality of images uploaded with clothing article listings, fashion websites are able to notify sellers of images that will reduce customer interest in an item. This will encourage sellers to improve aesthetic quality of their images, improving business
for both themselves and the fashion website.