treat bollean in cons of shape

onnx2kerastl/constant_layers.py

@@ -22,12 +22,23 @@ def convert_constant_of_shape(node, params, layers, lambda_func, node_name, kera
     value = params.get('value')
     if value is None:
         raise NotImplementedError("ConstantOfShape should have a value param")
+
     input_0 = layers[node.input[0]]
-    if not is_numpy(input_0) and not isinstance(input_0, list) and K.is_keras_tensor(
-        input_0):
-        layers[node_name] = tf.ones(layers[node.input[0]], dtype=tf.as_dtype(value.dtype)) * params['value']
+
+    if not is_numpy(input_0) and not isinstance(input_0, list) and K.is_keras_tensor(input_0):
+        # Boolean case
+        if value.dtype == np.bool_:
+            layers[node_name] = tf.fill(layers[node.input[0]], tf.constant(value.item(), dtype=tf.bool))
+        else:
+            # Non-boolean case
+            layers[node_name] = tf.ones(layers[node.input[0]], dtype=tf.as_dtype(value.dtype)) * value
     else:
-        layers[node_name] = np.ones(layers[node.input[0]], dtype=value.dtype) * params['value']
+        # Handle numpy inputs or non-Keras tensors
+        if value.dtype == np.bool_:
+            layers[node_name] = np.full(layers[node.input[0]], value.item(), dtype=bool)
+        else:
+            layers[node_name] = np.ones(layers[node.input[0]], dtype=value.dtype) * value
+
 
 
 def convert_one_hot(node, params, layers, lambda_func, node_name, keras_name):