Multimodel_setup / scripts /vit_adapter_test_results.json
Thiru1122's picture
feat(deepfake-detection): comprehensive cross-model evaluation, parity audits, and Phase 7 v2 reference preprocessing
151f388
Raw History Blame Contribute Delete
5.84 kB
[
{
"test_id": 1,
"name": "Real-image pipeline test",
"category": "VALID_IMAGE_PIPELINE",
"input": "C:\\Users\\user\\Desktop\\multi model connection\\benchmark\\test_images\\real.jpg",
"ground_truth_label": "UNKNOWN",
"expected_outcome": "SUCCESSFUL_INFERENCE",
"operational_status": "SUCCESSFUL_INFERENCE",
"domain_assessment": "NOT_EVALUATED: ground truth unknown",
"result": {
"detector_name": "ViTDeepfakeDetector",
"model_name": "dima806/deepfake_vs_real_image_detection (ViT-Base/16)",
"prediction": "REAL",
"fake_probability": 0.005097,
"real_probability": 0.994903,
"raw_logit": -5.273998,
"confidence": 0.994903,
"threshold": 0.5,
"device": "cpu",
"preprocessing": "ViTImageProcessor(Resize 224x224, Rescale 1/255, Normalize mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])",
"inference_time_ms": 8288.12,
"error": null,
"warning": "Vision Transformer trained for facial deepfake detection. Evaluates learned visual artifact representations; does not guarantee authenticity."
}
},
{
"test_id": 2,
"name": "AI-generated image",
"category": "VALID_AI_IMAGE",
"input": "C:\\Users\\user\\Desktop\\multi model connection\\benchmark\\test_images\\ai_generated.jpg",
"ground_truth_label": "FAKE",
"expected_outcome": "SUCCESSFUL_INFERENCE",
"operational_status": "SUCCESSFUL_INFERENCE",
"domain_assessment": "MATCH",
"result": {
"detector_name": "ViTDeepfakeDetector",
"model_name": "dima806/deepfake_vs_real_image_detection (ViT-Base/16)",
"prediction": "FAKE",
"fake_probability": 0.511827,
"real_probability": 0.488173,
"raw_logit": 0.047315,
"confidence": 0.511827,
"threshold": 0.5,
"device": "cpu",
"preprocessing": "ViTImageProcessor(Resize 224x224, Rescale 1/255, Normalize mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])",
"inference_time_ms": 673.04,
"error": null,
"warning": "Vision Transformer trained for facial deepfake detection. Evaluates learned visual artifact representations; does not guarantee authenticity."
}
},
{
"test_id": 3,
"name": "Manipulated-style image",
"category": "VALID_MANIPULATED_IMAGE",
"input": "C:\\Users\\user\\Desktop\\multi model connection\\benchmark\\test_images\\manipulated.jpg",
"ground_truth_label": "MANIPULATED",
"expected_outcome": "SUCCESSFUL_INFERENCE",
"operational_status": "SUCCESSFUL_INFERENCE",
"domain_assessment": "NOT_EVALUATED: ground truth is not a supported binary REAL/FAKE label",
"result": {
"detector_name": "ViTDeepfakeDetector",
"model_name": "dima806/deepfake_vs_real_image_detection (ViT-Base/16)",
"prediction": "REAL",
"fake_probability": 0.003638,
"real_probability": 0.996362,
"raw_logit": -5.612666,
"confidence": 0.996362,
"threshold": 0.5,
"device": "cpu",
"preprocessing": "ViTImageProcessor(Resize 224x224, Rescale 1/255, Normalize mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])",
"inference_time_ms": 793.58,
"error": null,
"warning": "Vision Transformer trained for facial deepfake detection. Evaluates learned visual artifact representations; does not guarantee authenticity."
}
},
{
"test_id": 4,
"name": "Corrupted image file",
"category": "CORRUPTED_FILE",
"input": "C:\\Users\\user\\Desktop\\multi model connection\\benchmark\\test_images\\corrupted_image.png",
"ground_truth_label": "CORRUPTED",
"expected_outcome": "INPUT_REJECTION",
"operational_status": "INPUT_REJECTION_VERIFIED",
"domain_assessment": "NOT_EVALUATED: inference/input error",
"result": {
"detector_name": "ViTDeepfakeDetector",
"model_name": "dima806/deepfake_vs_real_image_detection (ViT-Base/16)",
"prediction": "ERROR",
"fake_probability": null,
"real_probability": null,
"raw_logit": null,
"confidence": null,
"threshold": null,
"device": "cpu",
"preprocessing": "ViTImageProcessor(Resize 224x224, Rescale 1/255, Normalize mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])",
"inference_time_ms": null,
"error": "Invalid or corrupted image: cannot identify image file 'C:\\\\Users\\\\user\\\\Desktop\\\\multi model connection\\\\benchmark\\\\test_images\\\\corrupted_image.png'",
"warning": "Vision Transformer trained for facial deepfake detection. Evaluates learned visual artifact representations; does not guarantee authenticity."
}
},
{
"test_id": 5,
"name": "Plain text submitted as image",
"category": "UNSUPPORTED_INPUT",
"input": "C:\\Users\\user\\Desktop\\multi model connection\\benchmark\\test_images\\_temp_vit_invalid.txt",
"ground_truth_label": "INVALID_TYPE",
"expected_outcome": "INPUT_REJECTION",
"operational_status": "INPUT_REJECTION_VERIFIED",
"domain_assessment": "NOT_EVALUATED: inference/input error",
"result": {
"detector_name": "ViTDeepfakeDetector",
"model_name": "dima806/deepfake_vs_real_image_detection (ViT-Base/16)",
"prediction": "ERROR",
"fake_probability": null,
"real_probability": null,
"raw_logit": null,
"confidence": null,
"threshold": null,
"device": "cpu",
"preprocessing": "ViTImageProcessor(Resize 224x224, Rescale 1/255, Normalize mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])",
"inference_time_ms": null,
"error": "Invalid or corrupted image: cannot identify image file 'C:\\\\Users\\\\user\\\\Desktop\\\\multi model connection\\\\benchmark\\\\test_images\\\\_temp_vit_invalid.txt'",
"warning": "Vision Transformer trained for facial deepfake detection. Evaluates learned visual artifact representations; does not guarantee authenticity."
}
}
]