

906: How Prof. Jason Corso Solved Computer Vision’s Data Problem
74 snips Jul 18, 2025
Jason Corso, a Professor at the University of Michigan and co-founder of Voxel51, shares insights into revolutionizing computer vision. He discusses Voxel51’s powerful tool, Verified Auto-Labelling, which is transforming data quality in AI projects. The conversation explores the shift towards data-centric methodologies and the pivotal role of computer vision conferences in advancing research. Corso also highlights projects that merge AI with human-centric technology, enhancing daily tasks such as cooking and healthcare.
AI Snips
Chapters
Books
Transcript
Episode notes
Jason's Buffalo Architecture Story
- Jason Corso shared his connection to Buffalo's architecture and the AKG art museum.
- He once owned a historic house designed by local architect E.B. Green there.
Data Quality Drives Model Success
- Jason Corso found model performance depends more on the training dataset quality than on the choice of algorithm architecture.
- He observed that data quality crucially drives AI model success, especially in computer vision tasks like pedestrian avoidance.
Emphasize Data in ML Systems
- Data-centric machine learning requires equal attention to data quality as model development.
- Use tooling to analyze and improve data alongside models for better AI outcomes.