Skip to content
Back to skills

Dataset Annotation

ASecurity

"Pseudo-label image folders with GroundingDINO detections, optional

  • 247 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added September 8, 2026
toolspythonbashdebuggingapi

Works with

  • api

Security analysis

A100/100

Pro scans all 5 files and shows the line behind each finding

Scanned September 8, 2026

npx -y skills add VectorSpaceLab/AREX-Skill --skill dataset-annotation --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Dataset Annotation?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Dataset Annotation
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/vectorspacelab-dataset-annotation/badge)](https://www.skillsdirectory.com/skills/vectorspacelab-dataset-annotation)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: dataset-annotation
description: "Pseudo-label image folders with GroundingDINO detections, optional
  COCO export, and annotated-image review while keeping optional dataset tooling
  explicit."
disable-model-invocation: true
metadata:
  disco-role: operating
license: Apache 2.0
---

# Dataset Annotation

Use this sub-skill when the task is to pseudo-label a folder of images with GroundingDINO, inspect annotated review images, or export detections as a COCO-style dataset.

## Route first

- For one image, direct API inference, token spans, or prompt debugging, use `../inference/`.
- For COCO AP or benchmark evaluation, use `../evaluation/`.
- For Gradio, notebooks, Grounded-SAM, or image-editing integrations, use `../integrations/`.

## What this covers

- Folder pseudo-labeling with `image_directory`, `text_prompt`, `box_threshold`, `text_threshold`, `weights_path`, `config_path`, and optional `subsample` controls.
- Optional COCO export and annotated-image review outputs without launching a GUI by default.
- Safe handling for optional `fiftyone` and Typer-style workflow dependencies; the bundled helper uses `argparse`, not Typer.
- Conversion from GroundingDINO normalized `cxcywh` predictions into FiftyOne relative `xywh` detections.

## Bundled helper

From this sub-skill directory, start with:

```bash
python scripts/grounding_dino_pseudolabel.py --help
```

Minimal non-GUI COCO export example:

```bash
python scripts/grounding_dino_pseudolabel.py \
  --image-directory ./images \
  --text-prompt "bus . car ." \
  --config-path ./GroundingDINO_SwinT_OGC.py \
  --weights-path ./groundingdino_swint_ogc.pth \
  --export-coco \
  --draw-labels \
  --output-dir ./pseudolabel-output \
  --subsample 25
```

The helper never downloads checkpoints or datasets. Provide an existing config file and checkpoint, and pass `--view` only when a local FiftyOne GUI/server is intended.

## References

- Workflow and commands: `references/pseudolabel-workflow.md`
- Detection and export formats: `references/data-formats.md`
- Failure modes and recovery: `references/troubleshooting.md`

Files in this skill

  • SKILL.md2.1 KB
  • references/data-formats.md2.7 KB
  • references/pseudolabel-workflow.md4.1 KB
  • references/troubleshooting.md4 KB
  • scripts/grounding_dino_pseudolabel.py8.8 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.