Update app.py

app.py

@@ -96,6 +96,8 @@ char_to_num = layers.StringLookup(vocabulary=list(characters), mask_token=None)
 num_to_char = layers.StringLookup(
     vocabulary=char_to_num.get_vocabulary(), mask_token=None, invert=True
 )
+image_width = 128
+image_height = 32
 
 def distortion_free_resize(image, img_size):
   w, h = img_size
@@ -127,22 +129,20 @@ def distortion_free_resize(image, img_size):
           [0, 0],
       ],)
   image = tf.transpose(image, perm=[1,0,2])
-  image = tf.image.resize_with_crop_or_pad(image, target_height=128, target_width=32)
   image = tf.image.flip_left_right(image)
   return image
 
-def decode_batch_predictions(input_image):
-    img_size=(128, 32)
+def decode_batch_predictions(input_image, img_size=(image_width, image_height)):
     img_byte_array = io.BytesIO()
     input_image.save(img_byte_array, format='JPEG')  # Change the format as needed
     input_image = img_byte_array.getvalue()
+    
     input_image = tf.io.decode_image(input_image, channels=1,  dtype=tf.dtypes.uint8) 
     input_image = distortion_free_resize(input_image, img_size)
     input_image = tf.image.convert_image_dtype(input_image, tf.float32)/255.0
     
     pred = loaded_model.predict(input_image)
     input_len = np.ones(pred.shape[0]) * pred.shape[1]
-    # Use greedy search. For complex tasks, you can use beam search.
     results = keras.backend.ctc_decode(pred, input_length=input_len, greedy=True)[0][0][
         :, :max_len
     ]