bg-szy/TOP-SKILLS / skills/agent-skills/deepstream-import-vision-model
deepstream-import-vision-model
VerifiedNVIDIA DeepStream model-import guidance for bringing vision models from Hugging Face or NVIDIA NGC into DeepStream pipelines with export, TensorRT build, and benchmark steps.
| File | Kind | Size |
|---|---|---|
| BENCHMARK.md | markdown | 4.9 KB |
| CHANGELOG.md | markdown | 5.0 KB |
| README.md | markdown | 4.0 KB |
| SKILL.md | skill md | 16.6 KB |
| install.ps1 | script | 5.4 KB |
| install.sh | script | 6.3 KB |
| setup.sh | script | 3.6 KB |
| skill-card.md | markdown | 4.5 KB |
| skill.oms.sig | other | 16.1 KB |
README.md
View on GitHub# DeepStream Import Vision Model
Automated end-to-end pipeline: HuggingFace model → TensorRT engine → DeepStream
multi-stream benchmark → PDF report.
## Overview
This self-contained skill uses four phase-specific reference documents. Together they automate the full model bringup workflow for NVIDIA DeepStream, from
downloading a model on HuggingFace or NVIDIA NGC to a publication-ready benchmark report.
Supported input formats: ONNX (direct), SafeTensors (auto-exported via `torch.onnx.export`).
**Current scope:** object detection models only. Classification, segmentation, pose estimation, and other vision tasks are not yet supported — the pipeline fails fast if a non-detection architecture is detected in `config.json`.
## Prerequisites
**Host: only Docker + the NVIDIA driver.** Everything else runs **inside the DeepStream container** —
DeepStream, TensorRT/`trtexec`, the Python export venv (`torch`/`onnx`/`onnxruntime`), `wkhtmltopdf`,
`deepstream-app`/`gst-launch-1.0` — and is bootstrapped by `setup.sh`. Nothing is installed on the host.
Runs identically on Linux and **Windows** (Docker Desktop + WSL2 backend, required for GPU).
- **Docker** — Docker Desktop with the WSL2 backend on Windows; Docker Engine + NVIDIA Container
Toolkit on Linux.
- **NVIDIA GPU + driver** (on Windows, the WSL2 GPU driver — no host CUDA/TensorRT/DeepStream needed).
- `docker pull nvcr.io/nvidia/deepstream:9.1-triton-multiarch`, then run `setup.sh` through the container.
See **[references/windows.md](references/windows.md)** for the cross-platform run model and the
per-shell `docker run` mount token.
## Installation
```bash
bash <path-to-deepstream-import-vision-model>/install.sh --target <your-project-path>
```
Preview what will be installed first with `--dry-run`:
```bash
bash <path-to-deepstream-import-vision-model>/install.sh --target <your-project-path> --dry-run
```
Where `<path-to-deepstream-import-vision-model>` is the location of this skill in your repo, e.g.:
- In **team-mind-hub**: `team-skills/deepstream-sdk/deepstream-import-vision-model`
- In **ds-copilot**: `team-skills/deepstream-sdk/deepstream-import-vision-model` (same path)
The script copies the complete skill into the target project for Claude Code, Codex, and Cursor. Re-running it safely refreshes an existing installation.
## Usage
**Claude Code:**
```text
Use deepstream-import-vision-model to run this model: https://huggingface.co/onnx-community/yolov8n
```
**Codex:**
```text
Use $deepstream-import-vision-model to deploy and benchmark https://huggingface.co/onnx-community/yolov8n
```
**Cursor:**
```text
@deepstream-import-vision-model run this model: https://huggingface.co/onnx-community/yolov8n
```
The skill runs the full pipeline autonomously — no manual steps required.
## Pipeline Steps
| Step | Phase reference | Action |
|------|-----------|--------|
| 1–3 | `references/model-acquire.md` | Browse HF repo, download ONNX or export SafeTensors |
| 4–5 | `references/engine-build.md` | Build dynamic TRT engine, run trtexec benchmarks |
| 6–7 | `references/pipeline-run.md` | Custom bbox parser, DeepStream single + multi-stream |
| 8 | `references/report-generation.md` | 5 charts, HTML report, PDF |
## Output Structure
Per-model outputs are written to `models/<model_name>/` in your project:
```text
models/<model_name>/
model/ ONNX file(s)
parser/ Custom nvinfer bbox parser (.cpp, .so)
config/ nvinfer config, DS app config, labels.txt
scripts/ Model-specific run helpers
benchmarks/ TRT engines, trtexec logs
reports/ benchmark_report.md / .html / .pdf + charts/
samples/ Output videos, test frames, KITTI detections
```
## Files in this package
```text
deepstream-import-vision-model/
├── SKILL.md Top-level skill definition
├── README.md This file
├── references/ Phase-specific runbooks
├── scripts/ Utility scripts by pipeline phase
└── tests/ Installer and script regression tests
```
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deployment | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy. | Stars: 119k | Updated | Claude CodeCodex | Verified | Install |
| childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release. | Stars: 118k | Updated | Claude CodeCodex | Verified | Install |