ROS 2 TensorRT node turns Depth Anything V3 into real-time depth and point clouds

tom_doerr · x · 2026-07-29

A GitHub project adds a ROS 2 TensorRT node for Depth Anything V3, turning monocular camera images into real-time metric depth maps and PointCloud2 outputs.

What it does

Why it matters

The repo is framed as a practical robotics/perception integration rather than a model announcement, with a demo video and deployment-oriented docs for building, Docker, performance, and post-processing.

Original post →

More from Embodied

Embodied channel →