The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Understanding Cultural Style Trends with Computer Vision w/ Kavita Bala - #410
Sep 17, 2020
Kavita Bala, Dean of Computing and Information Science at Cornell University and research advisor at Facebook, dives into the fascinating world of computer vision and graphics. She discusses GrokStyle, a startup enhancing visual recognition for commerce on Facebook Marketplace. The conversation touches on how social media data is being leveraged to discover global style trends through projects like StreetStyle/GeoStyle. Kavita also highlights the importance of privacy-preserving techniques in technology, addressing cultural implications and innovations in the field.
AI Snips
Chapters
Transcript
Episode notes
Vision and Graphics Intertwined
- Computer graphics and computer vision are interconnected, like yin and yang.
- Graphics builds image models, while vision deciphers them, linked by human perception.
GrokStyle's Origin
- GrokStyle's origin traces back to material recognition research for a hypothetical house-cleaning robot.
- It evolved into fine-grained image recognition due to user demand on interior design sites.
Fine-Grained Recognition
- Fine-grained recognition uses Siamese networks to create embeddings, clustering similar items.
- Training involves correlating dissimilar-looking images of the same item, like catalog and real-world photos.