test_results.json

{
  "optimal_threshold": 0.1163,
  "performance": {
    "accuracy": 0.775,
    "precision": 0.7222,
    "recall": 0.8875,
    "f1": 0.7946,
    "model_agreement": 0.53
  },
  "datasets_tested": [
    "cifar10_vs_mnist",
    "cifar10_vs_fashion"
  ],
  "total_test_images": 800,
  "model_agreement_rate": 0.53,
  "recommendations": {
    "default_threshold": 0.1163,
    "high_precision": 0.5,
    "high_recall": 0.05
  }
}