The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Benchmarking Custom Computer Vision Services at Urban Outfitters with Tom Szumowski - TWiML Talk #247
Apr 3, 2019
Tom Szumowski, a Data Scientist at URBN, shares his insights on automating fashion product attribution using custom vision APIs. He discusses the transition from defense to retail, revealing unexpected connections between the two fields. Tom explains the challenges of building machine learning models for image recognition and the significance of data quality in model performance. He also compares various computer vision solutions, highlighting the effectiveness of their homegrown Fast.ai model. Listeners will gain a fascinating glimpse into the intersection of AI and e-commerce.
AI Snips
Chapters
Transcript
Episode notes
Career Shift
- Tom Szumowski transitioned from electrical engineering at Lockheed Martin, working on signal processing and cognitive electronic warfare, to data science at Urban Outfitters.
- This career shift involved applying similar underlying technologies in a new domain, focusing on product attribution in fashion retail.
Product Attribution
- Product attribution in fashion retail involves describing products with textual metadata, such as sleeve length, neckline, and color, for various business applications.
- These attributes are crucial for online navigation filters, search, planning, forecasting, and customer experience.
Attribution Process
- Product attribution at Urban Outfitters involves stages from buying to website listing, with manual enrichment of attributes.
- This manual process ensures accuracy and tailoring to trends, but automation is a long-term goal.