gt-audit v0.2.0

What's New

• Added generator and generator_version fields to JSON report output
• Reports now clearly identify they were generated by gt-audit

Report Format

{
  "generator": "gt-audit",
  "generator_version": "0.2.0",
  ...
}

gt-audit v0.1.0

gt-audit v0.1.0 - Initial Release

Fast ground truth label validation for object detection datasets.

Features

• YOLO model inference via ONNX Runtime
• Parallel image processing with rayon
• JSON report generation
• CI/CD ready with fail thresholds
• Support for YOLO format datasets

Usage

# Validate dataset with YOLO model
gt-audit validate /path/to/dataset --model model.onnx --confidence 0.25 --output report.json

# Show dataset info
gt-audit info /path/to/dataset

# CI/CD mode - fail if too many issues
gt-audit validate /path/to/dataset --model model.onnx --fail-on-high 10

Issue Types

• class_mismatch (High): Model detects different class than GT
• missing_label (Medium): Model detects object not in GT
• spurious_label (Low): GT has label model doesn't detect

Installation

cargo install --git https://github.com/ARTIFACTIQ/gt-audit

Or build from source:

git clone https://github.com/ARTIFACTIQ/gt-audit
cd gt-audit
cargo build --release