Upload 11 files

.dockerignore

@@ -0,0 +1,4 @@
+__pycache__
+*.pyc
+venv*
+.ipynb_checkpoints

Dockerfile

@@ -0,0 +1,17 @@
+FROM python:3.10-slim
+
+WORKDIR /app
+
+# system deps
+RUN apt-get update && apt-get install -y \
+    libgl1 \
+    && rm -rf /var/lib/apt/lists/*
+
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+COPY . .
+
+EXPOSE 7860
+
+CMD ["uvicorn", "Fastapi.main:app", "--host", "0.0.0.0", "--port", "7860"]

Fastapi/main.py

@@ -0,0 +1,73 @@
+from fastapi import FastAPI, UploadFile, File, HTTPException, Form
+import torch
+import os
+import uuid
+import numpy as np
+from PIL import Image
+from fastapi.middleware.cors import CORSMiddleware
+
+
+from inference.inference import (
+    convert_to_ela_image,
+    load_model
+)
+
+app = FastAPI(title="DeepFake Detection API")
+
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["*"],
+    allow_credentials=True,
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+# Load model once
+model, config, device = load_model()
+
+@app.post("/predict")
+async def predict(
+    file: UploadFile = File(...),
+    source: str = Form("upload")  # 👈 camera | upload
+):
+    if not file.content_type or not file.content_type.startswith("image/"):
+        raise HTTPException(status_code=400, detail="Uploaded file is not an image")
+
+    temp_filename = f"temp_{uuid.uuid4().hex}.jpg"
+
+    try:
+        contents = await file.read()
+        if not contents:
+            raise HTTPException(status_code=400, detail="Empty image file")
+
+        with open(temp_filename, "wb") as f:
+            f.write(contents)
+
+        # Validate image
+        try:
+            Image.open(temp_filename).verify()
+        except Exception:
+            raise HTTPException(status_code=400, detail="Invalid image")
+
+        # 🔥 CONDITIONAL PREPROCESSING
+        if source == "camera":
+            img = Image.open(temp_filename).convert("RGB").resize((128, 128))
+            img = torch.tensor(np.array(img) / 255.0, dtype=torch.float32)
+            img = img.permute(2, 0, 1).unsqueeze(0).to(device)
+        else:
+            img = convert_to_ela_image(temp_filename).to(device)
+
+        with torch.no_grad():
+            output = model(img)
+            probs = torch.softmax(output, dim=1)
+            pred = torch.argmax(probs, dim=1).item()
+            confidence = probs[0, pred].item()
+
+        return {
+            "prediction": config["class_mapping"][str(pred)],
+            "confidence": round(confidence * 100, 2)
+        }
+
+    finally:
+        if os.path.exists(temp_filename):
+            os.remove(temp_filename)

inference/inference.py

@@ -0,0 +1,106 @@
+import torch
+import torch.nn as nn
+import json
+import numpy as np
+import os
+from PIL import Image, ImageChops, ImageEnhance
+
+IMAGE_SIZE = (128, 128)
+
+# ---------------- ELA ----------------
+def convert_to_ela_image(image_path, quality=95):
+    original = Image.open(image_path).convert("RGB")
+
+    temp_path = "temp_ela.jpg"
+    original.save(temp_path, "JPEG", quality=quality)
+
+    compressed = Image.open(temp_path)
+    diff = ImageChops.difference(original, compressed)
+
+    extrema = diff.getextrema()
+    max_diff = max([ex[1] for ex in extrema])
+    scale = 255.0 / max_diff if max_diff != 0 else 1
+
+    diff = ImageEnhance.Brightness(diff).enhance(scale)
+    diff = diff.resize(IMAGE_SIZE)
+
+    ela = np.array(diff, dtype=np.float32) / 255.0
+    ela = torch.tensor(ela).permute(2, 0, 1).unsqueeze(0)
+
+    return ela
+
+
+# ---------------- MODEL ----------------
+class DeepFakeCNN(nn.Module):
+    def __init__(self):
+        super().__init__()
+
+        self.features = nn.Sequential(
+            nn.Conv2d(3, 32, kernel_size=5),
+            nn.ReLU(),
+            nn.MaxPool2d(2),
+
+            nn.Conv2d(32, 64, kernel_size=3),
+            nn.ReLU(),
+            nn.MaxPool2d(2),
+
+            nn.Conv2d(64, 128, kernel_size=3),
+            nn.ReLU(),
+            nn.MaxPool2d(2)
+        )
+
+        self._to_linear = None
+        self._get_flatten_size()
+
+        self.classifier = nn.Sequential(
+            nn.Linear(self._to_linear, 128),
+            nn.ReLU(),
+            nn.Dropout(0.4),
+            nn.Linear(128, 2)
+        )
+
+    def _get_flatten_size(self):
+        with torch.no_grad():
+            x = torch.zeros(1, 3, 128, 128)
+            x = self.features(x)
+            self._to_linear = x.view(1, -1).shape[1]
+
+    def forward(self, x):
+        x = self.features(x)
+        x = x.view(x.size(0), -1)
+        return self.classifier(x)
+
+
+# ---------------- LOAD MODEL ----------------
+BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+
+def load_model(
+    model_path=os.path.join(BASE_DIR, "models", "deepfake_cnn_2.pth"),
+    config_path=os.path.join(BASE_DIR, "models", "config_cnn_2.json")
+):
+    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+    with open(config_path, "r") as f:
+        config = json.load(f)
+
+    model = DeepFakeCNN().to(device)
+    model.load_state_dict(torch.load(model_path, map_location=device))
+    model.eval()
+
+    return model, config, device
+
+# Prediction Function
+
+def predict_image(image_path, model, config, device):
+    ela = convert_to_ela_image(image_path).to(device)
+
+    with torch.no_grad():
+        outputs = model(ela)
+        probs = torch.softmax(outputs, dim=1)
+        pred = torch.argmax(probs, dim=1).item()
+        confidence = probs[0, pred].item()
+
+    label = config["class_mapping"][str(pred)]
+
+    return label, confidence
+

models/config_cnn_2.json

@@ -0,0 +1,18 @@
+{
+    "model_name": "DeepfakeCNN_ELA",
+    "framework": "pytorch",
+    "input_size": [
+        128,
+        128,
+        3
+    ],
+    "ela_quality": 95,
+    "normalization": "divide_by_255",
+    "class_mapping": {
+        "0": "REAL",
+        "1": "FAKE"
+    },
+    "num_classes": 2,
+    "author": "Urvi",
+    "version": "1.0"
+}

models/deepfake_cnn_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0609adf33b9b8eda8633e2658f271f8a116ca2f948352586589e3dd1788a135b
+size 13230389

requirements.txt

@@ -0,0 +1,7 @@
+fastapi
+uvicorn
+torch
+torchvision
+pillow
+numpy
+python-multipart