Refactored prediction logic to display both FAKE and REAL class probabilities

app.py

@@ -113,13 +113,16 @@ def predict_deepfake(video_path, session_dir, cam_method):
             logits = output["logits"]
             probs = torch.softmax(logits, dim=1)  # (N, num_classes)
             probs_fake = probs[:, 0]  # (N,)
+            probs_real = probs[:, 1]
             avg_prob_fake = probs_fake.mean().item()
+            avg_prob_real = probs_real.mean().item()
 
         predicted_label = "FAKE" if avg_prob_fake >= 0.5 else "REAL"
-        confidence = avg_prob_fake if predicted_label == "FAKE" else 1 - avg_prob_fake
 
         result = (
-            f"**Prediction:** {predicted_label}\n" f"**Confidence:** {confidence:.2%}\n"
+            f"**Prediction:** {predicted_label}\n"
+            f"**Confidence FAKE:** {avg_prob_fake:.2%}\n"
+            f"**Confidence REAL:** {avg_prob_real:.2%}\n"
         )
 
         try: