How to Let Long-tail Guide Your 3D Annotation - December 9, 2026

Name: How to Let Long-tail Guide Your 3D Annotation - December 9, 2026
Start: 2026-12-09
End: 2026-12-09

Dec 09, 2026

9:00 AM - 10:00 AM PST

Online. Register for the Zoom!

Eric Hofesmann

Founding Machine Learning Engineer, Voxel51

About this event

Autonomous driving models rarely fail on the common case. They fail on the long-tail: the pedestrian stepping out from behind a parked truck or a construction zone that matches nothing in the training set. Each of those moments lives in synchronized multi-sensor data, made up of a LiDAR sweep, a ring of cameras, a timestamp, and a calibration. Fixing one means finding it, labeling it in 3D, and proving the next model handles it.

In this live session, we will take long-tail driving scenarios from discovery to retrained model, and make the case that 3D ground truth should be built where your data lives, not exported away from it.

Host

What you’ll learn:

How to find long-tail driving scenarios hiding in synchronized LiDAR and multi-camera data
How to label those scenarios in 3D where your data already lives, instead of exporting it to another tool
How to check that the next model actually handles the cases you fixed

Who this is for:

Autonomous vehicle and ADAS teams working with LiDAR, camera, and other sensor data
Machine learning engineers responsible for 3D ground truth and annotation quality
Data and annotation leads deciding where labeling should happen in their pipeline

You might also enjoy these events

Virtual

Building Composable Vision Workflows in Voxel51 - December 16, 2026

Join Harpreet Sahota for a live, open session on Voxel51, the visual AI data platform for computer vision and physical A...

Dec 16, 2026

Virtual

Visualizing and Debugging Robot Learning Data with Voxel51 and Hugging Face's LeRobot - November 19, 2026

Robot learning data doesn't look like the data that trains most vision-language models. It's temporal, low-dimensional i...

Nov 19, 2026