Update inference.py

inference.py

@@ -1,7 +1,13 @@
-import os
+# inference.py
+
 import tensorflow as tf
+import logging
 from preprocessing import VideoPreprocessor
 
+# Configure Logging
+logging.basicConfig(level=logging.INFO)
+logger = logging.getLogger(__name__)
+
 class LipReadingModel:
     def __init__(self, model_path='best_model_1_WER.keras'):
         # Initialize character mappings before loading the model
@@ -16,9 +22,10 @@ class LipReadingModel:
                 model_path,
                 custom_objects={'CTCLoss': self.CTCLoss}  # Include custom loss if needed
             )
-            print("Model loaded successfully.")
+            logger.info("Model loaded successfully.")
         except Exception as e:
-            print(f"Error loading model: {e}")
+            logger.error(f"Error loading model: {e}")
+            self.model = self.load_model()  # Fallback to building the model if loading fails
 
     @staticmethod
     def CTCLoss(y_true, y_pred):
@@ -69,24 +76,28 @@ class LipReadingModel:
         # Final dense layer
         model.add(tf.keras.layers.Dense(self.char_to_num.vocabulary_size() + 1, kernel_initializer='he_normal', activation='softmax'))
 
+        logger.info("Built the model architecture successfully.")
         return model
 
     def predict(self, normalized_frames):
         if normalized_frames is None or int(tf.size(normalized_frames)) == 0:
             return "No frames extracted from the video. Please ensure the video contains a clear view of the face and lips."
+
+        # Add batch dimension
+        frames = tf.expand_dims(normalized_frames, axis=0)  # Shape: (1, num_frames, 85, 85, 1)
         
-        frames = tf.expand_dims(normalized_frames, axis=0)  # Add batch dimension
+        # Perform prediction
         yhat = self.model.predict(frames, verbose=0)
         
         input_length = [yhat.shape[1]]  # batch size of 1
         # Perform CTC decoding
         decoded_tf = tf.keras.backend.ctc_decode(yhat, input_length=input_length, greedy=True)[0][0]
         decoded = decoded_tf.numpy().flatten()  # Convert to numpy array and flatten
-        
-        print("decoded type:", type(decoded))
-        print("decoded shape:", decoded.shape)
-        print("decoded contents:", decoded)
-        
+
+        logger.debug(f"decoded type: {type(decoded)}")
+        logger.debug(f"decoded shape: {decoded.shape}")
+        logger.debug(f"decoded contents: {decoded}")
+
         # Convert numerical predictions to characters
         prediction = ''.join(
             [
@@ -95,26 +106,25 @@ class LipReadingModel:
                 if int(num) != -1
             ]
         )
-        
-        return prediction
-
-
 
+        return prediction
 
-def predict_from_video(video_path=None, frames=None, model=None):
+def predict_from_video(video_path=None, frames=None, model=None, preprocessor=None):
     """
     Predicts the text from a video file or webcam frames using the provided model.
     Args:
         video_path (str, optional): Path to the video file. Defaults to None.
         frames (List[np.ndarray], optional): List of frames from webcam. Defaults to None.
         model (LipReadingModel, optional): An instance of the LipReadingModel. Defaults to None.
+        preprocessor (VideoPreprocessor, optional): An instance of the VideoPreprocessor. Defaults to None.
     Returns:
         str: Predicted text.
     """
     if model is None:
         model = LipReadingModel()
 
-    preprocessor = VideoPreprocessor()
+    if preprocessor is None:
+        preprocessor = VideoPreprocessor()
 
     if video_path:
         # Preprocess video from file