Agent skill · nvidia
foundationpose-setup
Install or repair the FoundationPose perception pipeline and build its FoundationStereo TensorRT engines. Use for SAM3/TAO dependency conflicts, CUDA library failures, and depth-engine shape or precision decisions.
What it needs
About 3k tokens when loaded.
What this skill does
FoundationPose perception pipeline setup Purpose Prepare the FoundationPose perception pipeline for depth, segmentation, and pose inference. Environment installation and engine construction belong here; dataset adaptation, inference, and pose evaluation belong to foundationpose-pipeline when that skill is installed. Requirements Work from the product checkout, not the installed skill directory. Find the user's checkout by checking for pyproject.toml (project foundationpose-perception-pipeline), tools/buildtaoengine.py, and config/defaults.yaml. If absent and setup was requested, clone the product URL above into the user's workspace and enter it. For advice-only requests, use the supplied diagnostics without cloning or installing anything. Commands below use paths relative to the product root; references/ links are relative to this skill. Read the checkout's README.md Requirements and Install sections for the matching revision. The supported stack requires Linux x8664, glibc >= 2.38, GLIBCXX3.4.31, NVIDIA driver >= 580, a CUDA toolkit >= 12.8 with nvcc, Python 3.12, uv, Git, Docker with GPU access, wget, and unzip. Start GPU sizing at 24 GB and measure the densest scene; 32 GB was tested. Budget depth-cache disk as roughly width height 4 3 bytes per scene, plus predictions and models. Keep sibling directories for sam3/, foundation-pose-inference-library/, and models/ beside the product checkout. models/ contains the deployable ONNX and engine, not FoundationStereo source. Instructions 1. Preflight before installing. Check glibc, GLIBCXX, driver, nvcc, tools, and Docker GPU access. An Ubuntu 22.04 host with glibc 2.35 cannot load the shipped FoundationPose library; report the unsupported runtime and stop setup there. Do not replace system libc or try to solve this with LDLIBRARYPATH. See installation. 2. Install into the product's Python 3.12 venv. Follow installation for uv, SAM3, the FoundationPose build, and TAO Deploy. …
How to use it
Reference it in AdaL, Claude Code, Cursor or any coding agent — nothing to install:
@skills nvidia/foundationpose-setup