NVIDIA AI Podcast Ep. 40: Using Deep Learning to Scan Your Shopping Basket
42 snips
Oct 25, 2017 Matt Scott, Co-founder and CTO of Malong Technologies, shares insights on their groundbreaking deep learning solutions for product recognition. He discusses the challenges of labeling data for effective AI training and the innovative use of weakly supervised learning. Scott highlights how their API-driven technology enhances retail experiences, enabling features like frictionless checkouts and accurate product tagging. His journey from New York to Beijing showcases the entrepreneurial spirit driving advancements in visual recognition technology and its potential to revolutionize shopping.
AI Snips
Chapters
Transcript
Episode notes
WebVision Challenge Win
- Malong Technologies won the WebVision challenge by achieving near-human image classification accuracy.
- Their key innovation is using weakly supervised learning, which leverages noisy, unlabeled web data, unlike ImageNet's meticulously labeled dataset.
ProductAI's Approach
- Malong's ProductAI focuses on general product recognition, identifying products and attributes from single images.
- Their 'mastermind' model utilizes attention mechanisms, focusing on key product features for accurate recognition, similar to human perception.
Demo for Jensen Huang
- Matt Scott demonstrated ProductAI to Jensen Huang, showcasing its ability to recognize unseen products.
- The demo highlighted the technology's potential for general product understanding.