The PHOTO2FIT project aims at augmenting the experience of online shopping for clothing items: given a real-world photo of a clothing item, e.g. taken on the street, PHOTO2FIT aims at finding that item, or a collection of very similar ones, in a pre-stored gallery of garments – such as an e-commerce catalog.
PHOTO2FIT models for the first time the situation where a person likes a clothing item that she/he sees when is walking around or reading a journal, and wants to have exactly that item, or a very similar one. The idea is to take a photo of that clothing item, submit it to a web application (possibly on a smartphone) and getting back a set of catalog items ranked by similarity, linked to the online shops where these items are sold.
To deal with these challenges, in PHOTO2FIT we will realize a prototype application where a user can submit a photo of a garment that she/he likes, obtaining a set of similar images ranked by similarity (from the most to the least similar one), referring to online products sold by an e-commerce platform.
This challenge fits well with the mission of the Joint Project, since the prototype will be the result of a tight connection between the academic partner, UNIVR, and the industrial partner, ACCORD SRL. In particular, UNIVR will study and test a machine learning algorithm based on deep learning to learn a similarity measure between street and shop photos, developing the machine intelligence at the core of the prototype. In addition, UNIVR will implement an interface of the prototype, so that novice users can use and evaluate it. ACCORD will focus on the commercial plausibility of the prototype, furnishing the data to analyze directly from a real e-commerce platform. ACCORD will also supervise the creation of the interface on top of the machine learning algorithm, and will help in judging the accuracy and the usability of the final prototype, thanks to a user study.