Apparel Clusterer

Purpose

This model is trained to recognize types of garments in a variety of settings like fashion shows, lookbooks, catalogues, fashion blogs, and social media posts.

Intended Use

Images of people who are shown 3/4 to full-length and facing forward or turned at an angle, with a range of backgrounds from simple, catalogue-style to busy street scenes.

Limitations

The model has a harder time with the following image types: Images that are cropped closely to a singular area of a person’s clothing (i.e. shoulders and head or only a torso), close-up images of a single garment, and images of garments not being worn by people.

Taxonomy

The taxonomy/ontology is meant to represent broad categories for detection purposes (that can later be refined in a workflow with a classification model). I.E. all types of tops – t-shirts, tank tops, blouses, turtlenecks, sweaters, etc – will be predicted as a top.