Isaac ROS DNN Stereo Depth

Type: Model / ROS Package Tags: NVIDIA, Isaac ROS, stereo depth, disparity, ESS, FoundationStereo, SGM, robotics, CUDA, TensorRT, NITROS Related: NVIDIA-Isaac-ROS, Isaac-ROS-DNN-Inference, Isaac-ROS-FoundationStereo, Isaac-ROS-nvblox, Isaac-ROS-Image-Pipeline, Isaac-ROS-NITROS, NVIDIA-Isaac-for-Manipulation, NVIDIA-Isaac-for-Mobility, NVIDIA-Jetson-Platform, TensorRT Sources: https://nvidia-isaac-ros.github.io/repositories_and_packages/isaac_ros_dnn_stereo_depth/index.html, https://nvidia-isaac-ros.github.io/concepts/stereo_depth/ess/index.html, https://nvidia-isaac-ros.github.io/concepts/stereo_depth/foundationstereo/index.html, https://nvidia-isaac-ros.github.io/releases/index.html Last Updated: 2026-04-29

Summary

Isaac ROS DNN Stereo Depth is NVIDIA’s Isaac ROS package family for deep stereo disparity and depth estimation. It covers model paths such as ESS and Isaac-ROS-FoundationStereo, comparing them against classical SGM-style stereo, and produces disparity maps that can be converted into depth and point clouds for robots.

Detail

Purpose

Robots need dense depth for obstacle avoidance, mapping, target tracking, and manipulation. DNN stereo methods learn matching costs and context from data, making them more robust in difficult regions such as textureless surfaces, reflections, occlusions, repetitive patterns, or minor calibration errors.

Key capabilities

  • Deep stereo disparity estimation from left/right stereo camera images.
  • ESS support for Efficient Semi-Supervised stereo disparity, optimized for real-time or compute-constrained uses.
  • Isaac-ROS-FoundationStereo support for higher-quality foundation-model stereo depth where real-time constraints are looser.
  • SGM comparison context for classical stereo disparity.
  • Use in NVIDIA-Isaac-for-Manipulation depth perception, including cuMotion-related workflows.
  • Depth input for Isaac-ROS-nvblox reconstruction and NVIDIA-Isaac-for-Mobility navigation.
  • NITROS acceleration for optimized disparity/image/tensor transport.

NVIDIA context

DNN Stereo Depth is the depth-perception layer between cameras and robot geometry. It feeds Isaac-ROS-nvblox maps, Isaac-ROS-cuMotion manipulation planning, Isaac-ROS-Visual-SLAM/mobility workflows, and Jetson-based perception pipelines.

Connections

Source Excerpts

  • NVIDIA docs describe DNN-based stereo depth as overcoming limitations of classical geometry-based stereo in difficult scenes.
  • The docs compare SGM, ESS, and FoundationStereo, with ESS and FoundationStereo providing smoother and more accurate disparity in NVIDIA examples.
  • ESS docs describe disparity prediction from stereo image pairs and converting disparity to depth or point clouds.

Resources