Hierarchical Multimodal Search for Data Curation in Autonomous Driving

About the Webinar

Finding specific driving scenarios takes more than searching for an object label. Engineers may need footage of a cyclist crossing at a junction while the vehicle slows. The search must capture the event and check that the motion condition holds at the same moment.

In this webinar, Philip Chen will show how Milvus supports autonomous driving data curation through semantic search, structured filters, and StructArray. We’ll explore a real-footage workflow for finding video windows, reviewing matches, and exporting candidate clips. A separate demo with synthetic driving scenes will compare scene-level search, object-level search, and combined results. You’ll see how to search specific details while keeping their scene context.

What You’ll Learn

• Search Across Scenes and Objects with Milvus — Compare scene-level search, object-level search, and combined results in a StructArray demo.
• Filter the Matching Moment — Apply motion conditions to the same video window that matches your query.
• Keep Related Data Together with StructArray — Store window embeddings and metadata within their parent clip. Learn when separate records make more sense.
• Follow a Data-Curation Workflow — Search real driving footage, inspect matching windows, and export candidate clips for review.
• Live Q&A — Ask Philip about multimodal search, Milvus StructArray, and your own data-curation use case.