An automated quality control tool for validating, scoring, and visualizing image annotation accuracy in computer vision datasets.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install image-annotation-qc
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install image-annotation-qc using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Image Annotation QC?
The Image Annotation QC skill is a specialized tool designed for developers and data scientists who need to ensure the highest quality for their AI training datasets. By integrating this component from the Openclaw Skills library, users can automatically audit bounding boxes and polygon segmentations across multiple industry-standard formats including COCO, YOLO, VOC, and LabelMe. The tool identifies common labeling mistakes and provides a weighted quality score to help teams decide if a dataset is ready for model training.
Beyond simple validation, the tool provides visual feedback by generating images that highlight specific errors with red-box annotations. This allows for rapid manual review and correction of datasets used in high-stakes environments like autonomous driving, industrial defect detection, and security monitoring.
Image Annotation QC Use Cases
- Validating autonomous driving datasets to ensure small objects and occlusions are correctly labeled.
- Auditing industrial inspection data to identify micro-defects that may have been missed or offset.
- Performing quality checks on security monitoring footage for consistent person and object tracking.
- Automated sampling of large-scale datasets to generate quality reports for project stakeholders.
How Image Annotation QC Works
- The user points the skill to a directory containing images and their corresponding annotation files.
- The system automatically detects the annotation format (COCO, YOLO, VOC, or LabelMe) to eliminate manual configuration.
- The skill applies scene-specific logic based on the selected domain (e.g., road, industrial, or security) to identify nuanced errors.
- It evaluates each annotation against criteria such as offsets, missing labels, duplicates, and incorrect class tags.
- A weighted quality score is calculated, grading the dataset from Excellent to Fail.
- The tool generates comprehensive reports in TXT, JSON, and Excel formats while creating visual overlays for all detected errors.
Image Annotation QC Setup
To use this tool from the Openclaw Skills collection, install the required Python dependencies and run the script from your terminal:
# Install dependencies
pip install Pillow openpyxl
# Basic usage with auto-detection
python3 scripts/qc_tool.py -i <image_dir> -a <annotation_dir>
# Advanced usage with Excel report and specific domain
python3 scripts/qc_tool.py -i ./images -a ./annotations -d industrial --formats txt json xlsx
Image Annotation QC Data Schema & Taxonomy
The skill organizes its output into a structured directory for easy integration into existing workflows:
| File/Folder | Purpose |
|---|---|
qc_report/ |
The root directory for all generated quality control data. |
qc_report.txt |
A human-readable summary of the annotation accuracy and error counts. |
qc_report.json |
A detailed, machine-readable file containing metadata for every detected error. |
qc_report.xlsx |
An optional formatted spreadsheet for project management and tracking. |
visual/ |
A sub-directory containing copies of images with red-box error visualizations. |
Image Annotation QC Advanced Features
- Scene-specific QC profiles for road safety, industrial inspection, and security monitoring.
- Automated format detection supporting COCO, YOLO, VOC, and LabelMe out of the box.
- Configurable error weighting to adjust quality scores based on project-specific priorities.
- Support for large-scale dataset sampling to provide rapid feedback on massive data volumes.
- Seamless integration as a Python module for inclusion in automated CI/CD data pipelines.
SKILL.md
Loading
Related Openclaw Skills
Dataset Splitter
A robust utility for automatically partitioning image datasets and annotations into training, validation, and testing subsets.
Daily Horoscope & Almanac
A specialized agent for querying traditional Chinese Almanac data, providing daily auspicious and inauspicious guidance.
Daily Fengshui Attire
A metaphysical dressing consultant that calculates the luckiest clothing colors based on the Chinese Earthly Branches and Five Elements of the day.
Bazi Pan - Eight Characters Fortune Telling
A sophisticated AI-driven charting tool for Bazi Eight Characters analysis using traditional Chinese metaphysics logic.
Image Cropper
A high-performance utility for extracting specific objects from image datasets based on standard bounding box annotation formats.
Image Deduplicator
A high-performance tool for detecting and managing duplicate or visually similar images using perceptual hashing.
METADATA
Featured*
