v1.0.0 — Initial Release

Fable Labeler v1.0.0

A local, offline, lightweight image annotation tool with built-in color verification, anomaly detection, and 3D RGB point cloud visualization.

Highlights

• Zero network dependency — runs entirely offline, no data leaves your machine
• RGB+HSV color verification — inspect what's actually inside your bounding boxes
• Triple-algorithm anomaly detection — Z-score + IQR + Mahalanobis distance with majority-vote consensus
• 3D RGB point cloud — visualize annotation color distributions in RGB space
• Multi-format export — COCO JSON, Pascal VOC XML, YOLO TXT, JSON, CSV

Requirements

• Python 3.9+
• Pillow >= 9.0
• numpy >= 1.21
• matplotlib >= 3.5

Quick Start

pip install -r requirements.txt
python main.py

What's Included

Module	Description
models/color_extractor.py	RGB/HSV extraction + anomaly detection engine
models/exporter.py	COCO / VOC / YOLO format exporters
models/point_cloud.py	3D RGB point cloud generation
models/annotation.py	Annotation + Project data model
ui/canvas_widget.py	Drawing canvas with zoom, pan, drag, resize
ui/main_window.py	Main window orchestration
tests/test_core.py	58 unit tests

Known Limitations

• Bounding box annotation only (no polygon/segmentation)
• Tkinter-based UI — functional but not pixel-perfect on high-DPI screens
• Anomaly detection is a statistical heuristic, not semantic correctness check
• Best suited for color-uniform object categories

License

MIT License — free for personal and commercial use.