Update app.py

app.py

@@ -12,15 +12,17 @@ processor = AutoImageProcessor.from_pretrained(model_name)
 model = SiglipForImageClassification.from_pretrained(model_name)
 model.eval()
 
-# Face detector
+# Load face detector
 face_detector = cv2.CascadeClassifier(cv2.data.haarcascades + "haarcascade_frontalface_default.xml")
 
 def analyze_deepfake(video_path):
     cap = cv2.VideoCapture(video_path)
     frame_preds = []
+    real_count = 0
+    fake_count = 0
     frame_count = 0
-    max_frames = 30   # ✅ Reduced for speed
-    frame_skip = 5    # ✅ Process every 5th frame
+    max_frames = 40
+    frame_skip = 5
 
     while True:
         ret, frame = cap.read()
@@ -32,60 +34,60 @@ def analyze_deepfake(video_path):
             continue
 
         gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
-        faces = face_detector.detectMultiScale(gray, scaleFactor=1.1, minNeighbors=5)
+        faces = face_detector.detectMultiScale(gray, 1.1, 4)
 
-        found = False
         for (x, y, w, h) in faces:
             face = frame[y:y+h, x:x+w]
             if face.size == 0:
                 continue
 
-            face_rgb = cv2.cvtColor(face, cv2.COLOR_BGR2RGB)
-            pil_img = Image.fromarray(face_rgb)
-            inputs = processor(images=pil_img, return_tensors="pt")
+            rgb_face = cv2.cvtColor(face, cv2.COLOR_BGR2RGB)
+            image = Image.fromarray(rgb_face)
+            inputs = processor(images=image, return_tensors="pt")
 
             with torch.no_grad():
                 logits = model(**inputs).logits
-                fake_prob = torch.softmax(logits, dim=-1)[0][1].item()
+                probs = torch.softmax(logits, dim=-1)[0]
+                fake_prob = probs[1].item()
+                frame_preds.append(fake_prob)
 
-            frame_preds.append(fake_prob)
-            found = True
-            break
-
-        if not found:
-            frame_preds.append(0.5)
+                if fake_prob > 0.6:
+                    fake_count += 1
+                else:
+                    real_count += 1
 
         frame_count += 1
 
     cap.release()
 
-    if frame_preds:
-        avg = np.mean(frame_preds)
-        verdict = "FAKE" if avg > 0.5 else "REAL"
-        result_text = f"✅ FINAL RESULT: **{verdict}** (confidence: {avg:.2f})"
+    # Decision logic
+    if real_count + fake_count == 0:
+        result = "❌ No faces detected. Try a clearer video."
     else:
-        result_text = "❌ No faces detected. Try a clearer video."
+        final_verdict = "FAKE" if fake_count > real_count else "REAL"
+        confidence = np.mean(frame_preds)
+        result = f"🎯 Result: **{final_verdict}** (Avg Confidence = {confidence:.2f}, Real = {real_count}, Fake = {fake_count})"
 
-    # Generate graph directly
+    # Graph
     fig, ax = plt.subplots(figsize=(6, 4))
-    ax.hist(frame_preds, bins=10, color="orange", edgecolor="black")
-    ax.set_title("Fake Confidence per Frame")
+    ax.hist(frame_preds, bins=10, color="green" if real_count > fake_count else "red", edgecolor="black")
+    ax.set_title("Fake Confidence per Face")
     ax.set_xlabel("Confidence (0 = Real, 1 = Fake)")
-    ax.set_ylabel("Frame Count")
+    ax.set_ylabel("Count")
     ax.grid(True)
 
-    return result_text, fig
+    return result, fig
 
 # Gradio UI
 demo = gr.Interface(
     fn=analyze_deepfake,
-    inputs=gr.Video(label="📤 Upload MP4 video"),
+    inputs=gr.Video(label="📤 Upload your .mp4 video (under 100MB)"),
     outputs=[
-        gr.Markdown(label="🧠 Analysis Result"),
-        gr.Plot(label="📈 Confidence Histogram")
+        gr.Markdown(label="📊 Deepfake Detection Result"),
+        gr.Plot(label="📈 Confidence Distribution")
     ],
-    title="🎭 Deepfake Video Detection (Fast)",
-    description="Upload a short MP4 video (under 60MB). This model will detect faces and classify each as REAL or FAKE based on frame analysis."
+    title="🎭 Deepfake Video Detector (Accurate & Fast)",
+    description="This model detects faces in video frames and classifies each as REAL or FAKE using a fine-tuned vision transformer."
 )
 
 demo.launch()