Clothing attribute detection (with Modelia.AI)

Supervisors: Veronika Cheplygina (ITU), Rene Haas (Modelia.AI)

This project is motivated by a real-life practical problem of Veronika: finding dresses or skirts with pockets. While online stores typically have metadata on the products, like length or type of fabric, surprisingly few have important properties “pockets” in the item description. This is even worse for stores with pre-owned items. Since the retailers won’t do it, we can try to find an alternative solution! The capabilities are all there, but there is not yet a practical yet versatile tool.

The project could involve steps such as:

• Scraping brands / clothes retailers for a specific clothing category
• Annotating images with specific attributes like pockets
• Using (off-the-shelf or trained) detectors to identify matching products
• Finding out whether the product is available in a physical store near you
• Analyzing pocket-friendliness of different brands over time

The project is co-advised by Modelia.AI, a company focusing on AI for various fashion-related tasks. Signing up for the project will require an interview with their product team before agreement.