OVERVIEW

Teams collect far more video than they can meaningfully label, and the events that matter most, including model failure cases, are usually buried in routine activity. This discussion brings together Thomas Tang from NVIDIA and Siddhartha Bal from iMerit to explore how teams find the scenarios worth labeling, prioritize annotation, generate targeted synthetic examples, and validate improvements on independent real world data. NVIDIA centers its Video Search and Summarization blueprint and Cosmos world foundation models, while iMerit brings annotation and expert review experience, including examples from autonomous vehicles.

What you will learn:

  • Why more video no longer means better coverage
  • How to tell a meaningful edge case from a merely rare one
  • Practical ways to surface the footage that matters
  • Where synthetic data helps and where real world evidence is essential
  • How expert review handles cases where the ground truth is unclear

REGISTRATION FORM