VPI Stereo Vision¶
Authors: Holoscan Team (NVIDIA)
Supported platforms: x86_64, aarch64
Last modified: March 18, 2025
Language: C++
Latest version: 1.0
Minimum Holoscan SDK version: 2.4.0
Tested Holoscan SDK versions: 2.4.0, 2.6.0
Contribution metric: Level 1 - Highly Reliable
Overview¶
Demo pipeline showing stereo disparity estimation using the Vision Programming Interface VPI.
Description¶
This pipeline takes video from a stereo camera and uses VPI's stereo disparity estimation algorithm. The input video and estimate disparity map are displayed using Holoviz.
The application will select accelerator backends if available (OFA, PVA and VIC). This demonstrates how VPI can be used to offload stereo disparity processing from the GPU on supported devices such as NVIDIA IGX, AGX, or NX platforms.
Input Video¶
Requires a V4L2 stereo camera, or recorded stereo video, and matching calibration data. By default, the application will share the same source data as stereo_vision.
Requirements¶
This demo requires VPI version 3.2 or greater. The included Dockerfile will install VPI and its dependencies for either an amd64 target (with discrete NVIDIA GPU), or arm64 target (NVIDIA IGX, AGX, or NX).
Build and Run Instructions¶
Using default video source (same as stereo_vision application):
./dev_container build_and_run vpi_stereo
./dev_container build_and_run vpi_stereo --run_args "--source v4l2"