Update app.py

app.py

@@ -8,23 +8,20 @@ import traceback
 from huggingface_hub import hf_hub_download
 
 # =======================================
-# Step 1: Download model from Hugging Face Hub
+# Step 1: Load TFLite model from Hugging Face Hub
 # =======================================
-keras_model = None
 interpreter, input_details, output_details = None, None, None
-use_tflite = False
 
 try:
-    # If you uploaded TFLite, use that:
-    # model_path = hf_hub_download("Shakeel401/Deepfake-Detector", "deepfake_model_select_ops.tflite")
+    model_path = hf_hub_download("Shakeel401/Deepfake-Detector", "deepfake_model_select_ops.tflite")
 
-    # But you uploaded .keras file:
-    model_path = hf_hub_download("Shakeel401/Deepfake-Detector", "deepfake_model.keras")
-
-    keras_model = tf.keras.models.load_model(model_path, compile=False)
-    print("✅ Loaded Keras model from HF Hub")
+    interpreter = tf.lite.Interpreter(model_path=model_path)
+    interpreter.allocate_tensors()
+    input_details = interpreter.get_input_details()
+    output_details = interpreter.get_output_details()
+    print("✅ Loaded TFLite model from HF Hub")
 except Exception as e:
-    print("❌ Failed to load model from HF Hub:", e)
+    print("❌ Failed to load TFLite model:", e)
 
 # =======================================
 # Step 2: OpenCV face detector
@@ -58,14 +55,13 @@ def preprocess_frames(frames, target_size=(299, 299)):
     return processed
 
 # =======================================
-# Step 4: Inference wrapper
+# Step 4: Inference wrapper (TFLite only)
 # =======================================
 def run_inference(processed):
     try:
-        if keras_model is not None:
-            return float(keras_model.predict(processed, verbose=0)[0][0])
-        else:
-            return 0.5  # fallback neutral
+        interpreter.set_tensor(input_details[0]['index'], processed)
+        interpreter.invoke()
+        return float(interpreter.get_tensor(output_details[0]['index'])[0][0])
     except Exception as e:
         print("❌ Inference failed:", e)
         return 0.5
@@ -76,8 +72,8 @@ def run_inference(processed):
 MAX_VIDEO_SIZE_MB = 50
 
 def predict_video_overlay(video_file):
-    if keras_model is None:
-        return "❌ No model available"
+    if interpreter is None:
+        return "❌ No TFLite model available"
 
     try:
         file_size_mb = os.path.getsize(video_file)/(1024*1024)
@@ -136,7 +132,7 @@ def predict_video_overlay(video_file):
 # =======================================
 # Step 6: Gradio App
 # =======================================
-title = "🎬 Deepfake Detection (Keras from Hugging Face Hub)"
+title = "🎬 Deepfake Detection (TFLite from Hugging Face Hub)"
 description = f"Upload MP4 (Max {MAX_VIDEO_SIZE_MB} MB). Detects Real/Fake and overlays predictions."
 
 gr.Interface(