Fixed checking length for symbolic tensor + make Model2onnx agnostic to the type of model

mltu/preprocessors.py

@@ -6,6 +6,8 @@
 import matplotlib
 import logging
 
+from typing import Type
+
 from . import Image
 from mltu.annotations.audio import Audio
 
@@ -18,7 +20,7 @@
 
 class ImageReader:
     """Read image from path and return image and label"""
-    def __init__(self, image_class: Image, log_level: int = logging.INFO, ) -> None:
+    def __init__(self, image_class: Type[Image], log_level: int = logging.INFO) -> None:
         self.logger = logging.getLogger(self.__class__.__name__)
         self.logger.setLevel(log_level)
         self._image_class = image_class

mltu/tensorflow/callbacks.py

@@ -1,5 +1,6 @@
 import os
 import tensorflow as tf
+from pathlib import Path
 from keras.callbacks import Callback
 
 import logging
@@ -14,7 +15,7 @@ def __init__(
         ) -> None:
         """ Converts the model to onnx format after training is finished.
         Args:
-            saved_model_path (str): Path to the saved .h5 model.
+            saved_model_path (str): Path to the saved model.
             metadata (dict, optional): Dictionary containing metadata to be added to the onnx model. Defaults to None.
             save_on_epoch_end (bool, optional): Save the onnx model on every epoch end. Defaults to False.
         """
@@ -35,35 +36,30 @@ def __init__(
 
     @staticmethod
     def model2onnx(model: tf.keras.Model, onnx_model_path: str):
-        try:
-            import tf2onnx
-
-            # convert the model to onnx format
-            tf2onnx.convert.from_keras(model, output_path=onnx_model_path)
+        import tf2onnx
 
-        except Exception as e:
-            print(e)
+        # convert the model to onnx format
+        # NOTE: see here for more info https://github.com/keras-team/keras/issues/18430
+        input_signature = [tf.TensorSpec(model.inputs[0].shape, model.inputs[0].dtype, name='digit')]
+        tf2onnx.convert.from_keras(model, input_signature=input_signature, opset=13, output_path=onnx_model_path)
 
     @staticmethod
     def include_metadata(onnx_model_path: str, metadata: dict=None):
-        try:
-            if metadata and isinstance(metadata, dict):
+        if metadata and isinstance(metadata, dict):
 
-                import onnx
-                # Load the ONNX model
-                onnx_model = onnx.load(onnx_model_path)
+            import onnx
+            # Load the ONNX model
+            onnx_model = onnx.load(onnx_model_path)
 
-                # Add the metadata dictionary to the model's metadata_props attribute
-                for key, value in metadata.items():
-                    meta = onnx_model.metadata_props.add()
-                    meta.key = key
-                    meta.value = str(value)
+            # Add the metadata dictionary to the model's metadata_props attribute
+            for key, value in metadata.items():
+                meta = onnx_model.metadata_props.add()
+                meta.key = key
+                meta.value = str(value)
 
-                # Save the modified ONNX model
-                onnx.save(onnx_model, onnx_model_path)
+            # Save the modified ONNX model
+            onnx.save(onnx_model, onnx_model_path)
 
-        except Exception as e:
-            print(e)  
 
     def on_epoch_end(self, epoch: int, logs: dict=None):
         """ Converts the model to onnx format on every epoch end. """
@@ -72,8 +68,8 @@ def on_epoch_end(self, epoch: int, logs: dict=None):
 
     def on_train_end(self, logs=None):
         """ Converts the model to onnx format after training is finished. """
-        self.model.load_weights(self.saved_model_path)
-        onnx_model_path = self.saved_model_path.replace(".h5", ".onnx")
+        self._model.load_weights(self.saved_model_path)
+        onnx_model_path = str(Path(self.saved_model_path).with_suffix('.onnx'))
         self.model2onnx(self.model, onnx_model_path)
         self.include_metadata(onnx_model_path, self.metadata)
 

mltu/tensorflow/transformer/utils.py

@@ -151,7 +151,7 @@ def update_state(self, y_true, y_pred, sample_weight=None):
         self.cer_accumulator.assign_add(tf.reduce_sum(distance))
 
         # Increment the batch_counter by the batch size
-        self.batch_counter.assign_add(len(y_true))
+        self.batch_counter.assign_add(y_true.shape[0])
 
     def result(self):
         """ Computes and returns the metric result.