Isaac ROS Image Pipeline
Type: Library / ROS Package Tags: NVIDIA, Isaac ROS, image pipeline, image processing, stereo, rectification, resize, CUDA, VPI, NITROS, Jetson Related: NVIDIA-Isaac-ROS, Isaac-ROS-NITROS, Isaac-ROS-DNN-Inference, Isaac-ROS-DNN-Stereo-Depth, Isaac-ROS-AprilTag, Isaac-ROS-FoundationStereo, NVIDIA-VPI, NVIDIA-Jetson-Platform, CV-CUDA Sources: https://nvidia-isaac-ros.github.io/repositories_and_packages/isaac_ros_image_pipeline/index.html, https://nvidia-isaac-ros.github.io/releases/index.html Last Updated: 2026-04-29
Summary
Isaac ROS Image Pipeline is NVIDIA’s accelerated ROS 2 image-processing metapackage for camera preprocessing and stereo image processing. It replaces CPU-heavy ROS image-pipeline stages with GPU and specialized-hardware accelerated nodes for operations such as rectification, resize, crop, mirroring, lens-distortion correction, color conversion, disparity, and point-cloud generation.
Detail
Purpose
Camera data often needs preprocessing before perception models, stereo depth, AprilTag detection, or mapping can use it. Isaac ROS Image Pipeline offloads common image-processing work from CPU to NVIDIA GPU and specialized hardware engines so robot CPUs remain available for control and coordination tasks.
Key capabilities
- Accelerated replacement path for ROS
image_pipelineworkflows on Jetson and NVIDIA GPU systems. - Image preprocessing including crop, resize, mirroring, lens-distortion correction, and color-space conversion.
- Stereo processing for disparity and point-cloud generation.
- Graph paths where rectified and resized images feed stereo disparity, DNN perception, AprilTag detection, and mapping.
- NITROS acceleration for optimized message formats.
- Current release support for ROS 2 Jazzy, JetPack 7.1, Jetson Thor, x86_64 NVIDIA GPUs, and DGX Spark signals.
NVIDIA context
Image Pipeline is the camera-preprocessing layer underneath many Isaac ROS perception pages. It feeds Isaac-ROS-DNN-Inference, Isaac-ROS-DNN-Stereo-Depth, Isaac-ROS-AprilTag, Isaac-ROS-FoundationStereo, and Isaac-ROS-nvblox with hardware-accelerated image transformations.
Connections
- NVIDIA-Isaac-ROS - parent ROS 2 package ecosystem.
- Isaac-ROS-NITROS - accelerated transport layer used by image pipeline nodes.
- Isaac-ROS-DNN-Inference - model inference graphs often require image preprocessing.
- Isaac-ROS-DNN-Stereo-Depth - stereo images need rectification and preprocessing before disparity estimation.
- Isaac-ROS-AprilTag - common graph uses rectify and resize before tag detection.
- Isaac-ROS-FoundationStereo - deep stereo model path that consumes stereo images.
- NVIDIA-VPI - adjacent NVIDIA vision-processing library and Jetson hardware-backend layer.
- NVIDIA-Jetson-Platform - edge deployment target for accelerated camera processing.
- CV-CUDA - adjacent NVIDIA GPU image-processing library for AI preprocessing.
Source Excerpts
- NVIDIA docs describe Isaac ROS Image Pipeline as a metapackage for image processing.
- The docs state that it is accelerated using GPU and specialized hardware engines.
- The package docs position it as a lower-CPU replacement for the standard ROS image_pipeline metapackage.