Main application file

app.py

@@ -0,0 +1,183 @@
+import os
+import time
+import requests
+import streamlit as st
+import torch
+from PIL import Image
+from torchvision import transforms
+from torchvision.transforms import InterpolationMode
+
+# ============================================================
+# Configuration
+# ============================================================
+
+MODEL_URL = (
+    "https://huggingface.co/neuralninja10/deepFakeWithCBAM/"
+    "resolve/main/deepFakeWithCBAM.pt"
+)
+
+MODEL_PATH = "deepFakeWithCBAM.pt"
+THRESHOLD = 0.68
+DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+# ============================================================
+# Page Config
+# ============================================================
+
+st.set_page_config(
+    page_title="DeepFake Detection Demo",
+    page_icon="🧠",
+    layout="centered",
+)
+
+# ============================================================
+# Secure Model Loader
+# ============================================================
+
+@st.cache_resource(show_spinner=False)
+def load_model():
+    token = os.environ.get("HF_TOKEN")
+    if token is None:
+        raise RuntimeError("HF_TOKEN is not set in Space secrets")
+
+    headers = {"Authorization": f"Bearer {token}"}
+
+    if not os.path.exists(MODEL_PATH):
+        with st.spinner("Initializing deepfake detection model..."):
+            response = requests.get(
+                MODEL_URL,
+                headers=headers,
+                stream=True,
+                timeout=60,
+            )
+            response.raise_for_status()
+            with open(MODEL_PATH, "wb") as f:
+                for chunk in response.iter_content(chunk_size=8192):
+                    f.write(chunk)
+
+    model = torch.jit.load(MODEL_PATH, map_location=DEVICE)
+    model.eval()
+    return model
+
+# ============================================================
+# Preprocessing
+# ============================================================
+
+_transform = transforms.Compose([
+    transforms.Resize((256, 256), interpolation=InterpolationMode.BILINEAR),
+    transforms.ToTensor(),
+    transforms.Normalize(
+        mean=[0.485, 0.456, 0.406],
+        std=[0.229, 0.224, 0.225],
+    ),
+])
+
+def preprocess_image(image: Image.Image) -> torch.Tensor:
+    return _transform(image).unsqueeze(0)
+
+# ============================================================
+# Inference
+# ============================================================
+
+def run_inference(model, image: Image.Image):
+    tensor = preprocess_image(image).to(DEVICE)
+
+    start = time.time()
+    with torch.no_grad():
+        logits = model(tensor)
+        prob = torch.sigmoid(logits).item()
+    latency_ms = (time.time() - start) * 1000
+
+    is_real = prob > THRESHOLD
+    confidence = prob if is_real else (1 - prob)
+
+    return {
+        "prediction": "Real" if is_real else "Fake",
+        "confidence": confidence,
+        "real_prob": prob,
+        "fake_prob": 1 - prob,
+        "latency_ms": latency_ms,
+    }
+
+# ============================================================
+# UI
+# ============================================================
+
+def main():
+    st.title("🧠 DeepFake Detection")
+    st.markdown(
+        """
+        Upload a facial image to determine whether it is **Real** or **AI-Generated**.
+        This demo runs entirely on **CPU** using a TorchScript model.
+        """
+    )
+
+    try:
+        model = load_model()
+    except Exception as e:
+        st.error("❌ Failed to load the model.")
+        st.exception(e)
+        return
+
+    uploaded_file = st.file_uploader(
+        "Upload an image (JPG / PNG)",
+        type=["jpg", "jpeg", "png"],
+        accept_multiple_files=False,
+    )
+
+    if uploaded_file:
+        image = Image.open(uploaded_file).convert("RGB")
+        st.image(image, caption="Uploaded Image", use_container_width=True)
+
+        if st.button("Run DeepFake Detection"):
+            with st.spinner("Running inference..."):
+                result = run_inference(model, image)
+
+            st.divider()
+            st.subheader("Detection Result")
+
+            if result["prediction"] == "Real":
+                st.success("✅ Real Face Detected")
+            else:
+                st.error("❌ Deepfake Detected")
+
+            st.metric(
+                label="Confidence Score",
+                value=f"{result['confidence']:.2%}",
+            )
+
+            st.caption(
+                f"Inference latency: {result['latency_ms']:.1f} ms (CPU)"
+            )
+
+            with st.expander("Detailed Probabilities"):
+                st.write(f"Real Probability: {result['real_prob']:.4f}")
+                st.write(f"Fake Probability: {result['fake_prob']:.4f}")
+
+    st.divider()
+
+    st.caption(
+        """
+        ⚠️ **Disclaimer**  
+        This system is provided for research and demonstration purposes only.
+        Predictions may be incorrect and should not be used as the sole basis
+        for real-world decisions.
+        """
+    )
+
+    with st.expander("Model Information"):
+        st.markdown(
+            """
+            - **Architecture:** EfficientNet + CBAM  
+            - **Input Resolution:** 256×256  
+            - **Runtime:** CPU (TorchScript)  
+            - **Threshold:** 0.68  
+            - **Known Limitations:**  
+              - Heavy compression artifacts  
+              - Extreme lighting conditions  
+              - Occluded or profile faces  
+            """
+        )
+
+if __name__ == "__main__":
+    main()