Upload folder using huggingface_hub

README.md

@@ -1,3 +1,63 @@
----
-license: mit
----
+# DeepFake Detector V14 🎯
+
+**Weighted ensemble of V13 models**
+
+## Overview
+
+V14 is a weighted ensemble that combines the three V13 models:
+- Model 13.1 (ConvNeXt-Large): Weight 0.25
+- Model 13.2 (ViT-Large): Weight 0.35
+- Model 13.3 (Swin-Large): Weight 0.40
+
+**Expected F1 Score**: 0.9361
+
+## Model Files
+
+- `model_1.safetensors` - ConvNeXt-Large (788 MB)
+- `model_2.safetensors` - ViT-Large (1220 MB)
+- `model_3.safetensors` - Swin-Large (783 MB)
+- `ensemble.pth` - Ensemble wrapper weights
+- `config.json` - Configuration
+- `inference_example.py` - Usage example
+
+## Quick Start
+
+```python
+# See inference_example.py for complete code
+# Load all 3 models, run predictions, compute weighted average
+```
+
+## Performance
+
+| Model | Backbone | F1 Score | Ensemble Weight |
+|-------|----------|----------|-----------------|
+| Model 13.1 | ConvNeXt-Large | 0.8971 | 0.25 |
+| Model 13.2 | ViT-Large | 0.9382 | 0.35 |
+| Model 13.3 | Swin-Large | 0.9586 | 0.40 |
+
+**Weighted Average F1**: 0.9361
+
+## Requirements
+
+```
+torch>=2.0.0
+timm>=0.9.0
+torchvision>=0.15.0
+safetensors
+pillow
+```
+
+## Citation
+
+```bibtex
+@model{v14-deepfake-detector,
+  title={DeepFake Detector V14},
+  author={Ash},
+  year={2024},
+  publisher={Hugging Face}
+}
+```
+
+## Predecessor
+
+Built on: [`ash12321/deepfake-detector-v13`](https://huggingface.co/ash12321/deepfake-detector-v13)

config.json

@@ -0,0 +1,67 @@
+{
+  "model_name": "DeepFake Detector V14",
+  "version": "14.0",
+  "architecture": "Weighted Ensemble",
+  "total_parameters": "699M",
+  "description": "Ensemble of 3 V13 models with weighted averaging",
+  "models": [
+    {
+      "id": 1,
+      "name": "Model 13.1",
+      "backbone": "convnext_large",
+      "parameters": "198M",
+      "dropout": 0.3,
+      "f1_score": 0.8971,
+      "ensemble_weight": 0.25,
+      "file": "model_1.safetensors"
+    },
+    {
+      "id": 2,
+      "name": "Model 13.2",
+      "backbone": "vit_large_patch16_224",
+      "parameters": "304M",
+      "dropout": 0.35,
+      "f1_score": 0.9382,
+      "ensemble_weight": 0.35,
+      "file": "model_2.safetensors"
+    },
+    {
+      "id": 3,
+      "name": "Model 13.3",
+      "backbone": "swin_large_patch4_window7_224",
+      "parameters": "197M",
+      "dropout": 0.3,
+      "f1_score": 0.9586,
+      "ensemble_weight": 0.4,
+      "file": "model_3.safetensors"
+    }
+  ],
+  "ensemble": {
+    "method": "weighted_average",
+    "weights": [
+      0.25,
+      0.35,
+      0.4
+    ],
+    "expected_f1": 0.9360850000000001
+  },
+  "usage": {
+    "preprocessing": {
+      "image_size": 224,
+      "normalization": {
+        "mean": [
+          0.485,
+          0.456,
+          0.406
+        ],
+        "std": [
+          0.229,
+          0.224,
+          0.225
+        ]
+      }
+    },
+    "inference": "Load all 3 models, get predictions, compute weighted average"
+  },
+  "predecessor": "ash12321/deepfake-detector-v13"
+}

ensemble.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:56c0ce099d756d0f653facc2b96128c31388a7e06e5eaa5a54ce554e4889f393
+size 2787760605

inference_example.py

@@ -0,0 +1,74 @@
+import torch
+import timm
+from torchvision import transforms
+from PIL import Image
+from safetensors.torch import load_file
+
+class DeepfakeDetector(torch.nn.Module):
+    def __init__(self, backbone_name, dropout=0.3):
+        super().__init__()
+        self.backbone = timm.create_model(backbone_name, pretrained=False, num_classes=0)
+        
+        if hasattr(self.backbone, 'num_features'):
+            feat_dim = self.backbone.num_features
+        else:
+            with torch.no_grad():
+                feat_dim = self.backbone(torch.randn(1, 3, 224, 224)).shape[1]
+        
+        self.classifier = torch.nn.Sequential(
+            torch.nn.Linear(feat_dim, 512),
+            torch.nn.BatchNorm1d(512),
+            torch.nn.GELU(),
+            torch.nn.Dropout(dropout),
+            torch.nn.Linear(512, 128),
+            torch.nn.BatchNorm1d(128),
+            torch.nn.GELU(),
+            torch.nn.Dropout(dropout * 0.5),
+            torch.nn.Linear(128, 1)
+        )
+    
+    def forward(self, x):
+        features = self.backbone(x)
+        return self.classifier(features).squeeze(-1)
+
+# Load models
+device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+
+configs = [
+    ('convnext_large', 0.3, 'model_1.safetensors', 0.25),
+    ('vit_large_patch16_224', 0.35, 'model_2.safetensors', 0.35),
+    ('swin_large_patch4_window7_224', 0.3, 'model_3.safetensors', 0.40)
+]
+
+models = []
+for backbone, dropout, filename, weight in configs:
+    model = DeepfakeDetector(backbone, dropout)
+    state_dict = load_file(filename)
+    model.load_state_dict(state_dict)
+    model = model.to(device)
+    model.eval()
+    models.append((model, weight))
+
+# Preprocess image
+transform = transforms.Compose([
+    transforms.Resize((224, 224)),
+    transforms.ToTensor(),
+    transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
+])
+
+# Predict
+image = Image.open('test.jpg').convert('RGB')
+input_tensor = transform(image).unsqueeze(0).to(device)
+
+with torch.no_grad():
+    predictions = []
+    for model, weight in models:
+        logits = model(input_tensor)
+        prob = torch.sigmoid(logits).item()
+        predictions.append(prob * weight)
+    
+    final_prob = sum(predictions)
+    prediction = 'FAKE' if final_prob > 0.5 else 'REAL'
+
+print(f"Prediction: {prediction}")
+print(f"Confidence: {final_prob:.2%}")

model_1.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f70541704e8eba1910990469e1a6f9d8a1badc451b2a4d2909170ee53ba45c9
+size 788381444

model_2.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3878f20c6949953030d9294132f4b333a5d9a4349a3fa91420270ec5f7a8ad8b
+size 1215611244

model_3.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:db558427cfd28173170921b45a0c8f869e122d6e46aff1abd83d55ce6a80f8b8
+size 783441332