Skip to content

Simet

Simet is a modular toolkit for evaluating synthetic images against real ones with reproducible metrics and threshold-based restraints.

pip install simet

Key features

  • Load images from folders or built-in datasets
  • Transform for feature extraction (Inception-v3 provided)
  • Extract and cache features automatically
  • Measure quality (FID, Precision/Recall, ROC-AUC)
  • Gate results with simple threshold rules

Why Simet?

  • 🔁 Fast iteration — cache features and tweak thresholds freely
  • ⚙️ Extensible — plug custom providers, transforms, metrics, restraints
  • 🚀 Scalable — FAISS IVF, batching, GPU/AMP, subsampling
  • 🧩 Deterministic — consistent seeds & reproducible logs

Typical workflow

  1. Point to real and synthetic image folders
  2. Pick metrics
  3. Optionally add restraints
  4. Run via CLI or Python API
  5. Inspect reports and JSON logs

→ Continue to Getting Started to run your first evaluation.