Publish DeepVerify V1 Model Bundle

README.md

@@ -0,0 +1,34 @@
+---
+tags:
+- deepfake-detection
+- computer-vision
+- efficientnet
+- forensics
+---
+
+# DeepVerify V1 Models
+
+DeepVerify is an AI-generated/manipulated media detection project. 
+This repository contains the V1 production model bundle, utilizing a 60/40 spatial-frequency probability ensemble.
+
+## Models Included
+
+1. **Spatial Model (EXP-001)**
+   - Architecture: EfficientNet-B0
+   - Representation: Spatial RGB
+   - Weight: 60% of ensemble
+
+2. **Frequency Model (EXP-002)**
+   - Architecture: EfficientNet-B0
+   - Representation: Frequency-domain log-magnitude FFT
+   - Weight: 40% of ensemble
+
+## Input & Output
+- **Expected Input**: 224x224 RGB image
+- **Output Interpretation**: Both models predict `P(REAL)`. The final ensemble score is `0.60 * Spatial_P(REAL) + 0.40 * Frequency_P(REAL)`. A score `>= 0.5` is considered `LIKELY_AUTHENTIC`, while `< 0.5` is considered `LIKELY_AI_GENERATED`.
+
+## Limitations
+- **Dataset Limitations**: Trained exclusively on the 140k Real and Fake Faces v1.0 benchmark. Does not guarantee out-of-distribution generalization to all generative models or non-face semantics.
+- **Model Limitations**: This model is **not** a universal deepfake detector. It is highly specific to the domain it was trained on.
+- **Confidence**: The probability score is a statistical confidence metric, **not** absolute ground truth.
+- **Explainability**: Spatial Grad-CAM visual evidence indicates regions of model activation/saliency. It is **not** causal proof of manipulation.

frequency/metadata.json

@@ -0,0 +1,22 @@
+{
+  "experiment_id": "EXP-002",
+  "run_id": "run_20261008_080128_559d2631",
+  "architecture": "EfficientNet-B0",
+  "pretrained_initialization": "ImageNet",
+  "input_size": [
+    3,
+    224,
+    224
+  ],
+  "preprocessing_representation": "Frequency-domain log-magnitude FFT",
+  "class_convention": "0=REAL, 1=FAKE",
+  "probability_convention": "P(REAL)",
+  "dataset_version": "140k-real-and-fake-faces-v1.0",
+  "checkpoint_filename": "model_best.pth",
+  "sha256": "507f9d6cf9aadc633d336f236df140555f4847fdc29697a8fc576a3fe3aa4c52",
+  "relevant_metrics": {
+    "test_accuracy": "81.73%",
+    "test_roc_auc": "0.9014"
+  },
+  "known_limitations": "Trained on 140k face benchmark. Not a universal deepfake detector."
+}

frequency/model_best.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:507f9d6cf9aadc633d336f236df140555f4847fdc29697a8fc576a3fe3aa4c52
+size 48606264

manifest.json

@@ -0,0 +1,21 @@
+{
+  "model_bundle_version": "v1.0.0",
+  "ensemble_weights": {
+    "spatial": 0.6,
+    "frequency": 0.4
+  },
+  "models": {
+    "spatial": {
+      "experiment_id": "EXP-001",
+      "run_id": "run_20261007_141902_9e04eeda",
+      "checkpoint_filename": "spatial/model_best.pth",
+      "sha256": "b05a302c9283783fdc7352abe967baa4e4ab278028a2ba08cf40bd290e521d03"
+    },
+    "frequency": {
+      "experiment_id": "EXP-002",
+      "run_id": "run_20261008_080128_559d2631",
+      "checkpoint_filename": "frequency/model_best.pth",
+      "sha256": "507f9d6cf9aadc633d336f236df140555f4847fdc29697a8fc576a3fe3aa4c52"
+    }
+  }
+}

spatial/metadata.json

@@ -0,0 +1,22 @@
+{
+  "experiment_id": "EXP-001",
+  "run_id": "run_20261007_141902_9e04eeda",
+  "architecture": "EfficientNet-B0",
+  "pretrained_initialization": "ImageNet",
+  "input_size": [
+    3,
+    224,
+    224
+  ],
+  "preprocessing_representation": "Spatial RGB",
+  "class_convention": "0=REAL, 1=FAKE",
+  "probability_convention": "P(REAL)",
+  "dataset_version": "140k-real-and-fake-faces-v1.0",
+  "checkpoint_filename": "model_best.pth",
+  "sha256": "b05a302c9283783fdc7352abe967baa4e4ab278028a2ba08cf40bd290e521d03",
+  "relevant_metrics": {
+    "test_accuracy": "99.895%",
+    "test_roc_auc": "0.999995"
+  },
+  "known_limitations": "Trained on 140k face benchmark. Not a universal deepfake detector."
+}

spatial/model_best.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b05a302c9283783fdc7352abe967baa4e4ab278028a2ba08cf40bd290e521d03
+size 48606264