# deepstream-import-vision-model

This skill takes an object-detection model from HuggingFace or NGC through a full NVIDIA DeepStream pipeline: ONNX, TRT engine, bbox parser, benchmark, PDF report.

- skill · `type=skill` · Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, OpenCode · 2300 installs · ★ 3735
- Author: Tushar Khinvasara
- License: not stated · policy=link-only · description only; the file stays with the author
- Source: https://github.com/NVIDIA/skills/tree/HEAD/skills/deepstream-import-vision-model/SKILL.md
- Updated: 2026-10-23
- HTML: https://affpapa.org/en/nevibecode/skills/deepstream-import-vision-model

## What it does

The skill takes a supported object-detection model and carries it all the way into a running DeepStream pipeline. It starts by offering two choices: the validated `PekingU/rtdetr_r50vd` from Hugging Face, or your own model — either a Hugging Face ID in `organization/model` form or an NGC catalog URL with its version. Classification, segmentation and other architectures get rejected after `config.json` is inspected, and the skill will not quietly swap in a different model.

From there it runs eight steps: download ONNX or export from SafeTensors, build a dynamic TRT engine with `trtexec` runs at BS=1 and BS=MAX_BS, generate a custom bbox parser for `nvinfer`, benchmark single-stream and multi-stream, then produce a report with charts as HTML and PDF. Everything executes inside the `nvcr.io/nvidia/deepstream:9.1-triton-multiarch` container, so the host needs only Docker and the NVIDIA driver — no python, torch or `trtexec` to install. On Windows you need Docker Desktop with the WSL2 backend, otherwise `--gpus` will not work.

Before the first run, the skill bootstraps the environment in-container via `setup.sh` and installs `wkhtmltopdf`, then checks GPU, venv and `trtexec` with `preflight.sh` — proceed only on PASS. Artifacts land under `models/{model_name}/` in a fixed layout, log and engine filenames are hard-coded, and a KITTI check gates the multi-stream step: zero frames or a detection rate under 90% stops the run.

## Requirements
- Works with: Docker

## Install

### Claude Code (native platform)

1. **Skills CLI (Vercel)**
```bash
npx skills add NVIDIA/skills --skill deepstream-import-vision-model -a claude-code
```

2. **Manually: put the files in place**
- File: `~/.claude/skills/deepstream-import-vision-model/SKILL.md` — Copy the whole skill folder; for one project use .claude/skills/deepstream-import-vision-model/

### Codex

1. **Skills CLI (Vercel)**
```bash
npx skills add NVIDIA/skills --skill deepstream-import-vision-model -a codex
```

2. **Manually: put the files in place**
- File: `.agents/skills/deepstream-import-vision-model/SKILL.md` — For all projects use ~/.agents/skills/

### Cursor

1. **Skills CLI (Vercel)**
```bash
npx skills add NVIDIA/skills --skill deepstream-import-vision-model -a cursor
```

2. **Manually: put the files in place**
- File: `.cursor/skills/deepstream-import-vision-model/SKILL.md` — Cursor also reads .agents/skills/ and .claude/skills/

### GitHub Copilot

1. **GitHub CLI**
```bash
gh skill install NVIDIA/skills deepstream-import-vision-model
```

2. **Manually: put the files in place**
- File: `.github/skills/deepstream-import-vision-model/SKILL.md`

### Gemini CLI

1. **Skills CLI (Vercel)**
```bash
npx skills add NVIDIA/skills --skill deepstream-import-vision-model -a gemini-cli
```

2. **Platform CLI**
```bash
gemini skills install https://github.com/NVIDIA/skills --path skills/deepstream-import-vision-model
```

### Windsurf (Devin Desktop)

**Manually: put the files in place**
- File: `.devin/skills/deepstream-import-vision-model/SKILL.md` — The legacy .windsurf/skills/ path still works

### OpenCode

**Skills CLI (Vercel)**
```bash
npx skills add NVIDIA/skills --skill deepstream-import-vision-model -a opencode
```

### Claude.ai

**Claude.ai**
- Zip the folder and upload it in Settings → Capabilities → Skills

Where a file is given instead of a command, copy the files from the source to that path: https://github.com/NVIDIA/skills/tree/HEAD/skills/deepstream-import-vision-model/SKILL.md

## When to use it
- pull an object-detection ONNX model from Hugging Face or NGC
- build a dynamic TRT engine for a given batch size
- write a bbox parser for nvinfer that matches the model
- benchmark a pipeline across several video streams
- produce a PDF report with charts from the measurements

## Example prompts
- Take PekingU/rtdetr_r50vd and build its TRT engine for DeepStream
- Import this model from NGC into a DeepStream pipeline and benchmark it on four streams
- Do a dry run: walk through the import steps without downloading or building anything

## Similar
- [senior-computer-vision](https://affpapa.org/en/nevibecode/skills/senior-computer-vision.md) — skill · 1152 installs
- [model-deployment](https://affpapa.org/en/nevibecode/skills/model-deployment.md) — skill · 379 installs
- [gpu-server-management](https://affpapa.org/en/nevibecode/skills/gpu-server-management.md) — skill · 331 installs
- [debug-openshell-cluster](https://affpapa.org/en/nevibecode/skills/debug-openshell-cluster.md) — skill · 474 installs
- [fal-vision](https://affpapa.org/en/nevibecode/skills/fal-vision.md) — skill · 2700 installs
- [Computer Vision](https://affpapa.org/en/nevibecode/skills/computer-vision.md) — skill · 499 installs

From the same set: [nvidia/skills](https://affpapa.org/en/nevibecode/by/nvidia/skills.md) — 178 cards

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md