fixing the conv auto-pads

README.md

@@ -29,7 +29,7 @@ pip install git+https://github.com/microsoft/onnxconverter-common
 pip install git+https://github.com/onnx/keras-onnx
 ```
 Before running the converter, please notice that tensorflow has to be installed in your python environment,
-you can choose **tensorflow** package(CPU version) or **tensorflow-gpu**(GPU version)
+you can choose **tensorflow**/**tensorflow-cpu** package(CPU version) or **tensorflow-gpu**(GPU version)
 
 # Notes
 Keras2ONNX supports the new Keras subclassing model which was introduced in tensorflow 2.0 since the version **1.6.5**. Some typical subclassing models like [huggingface/transformers](https://github.com/huggingface/transformers) have been converted into ONNX and validated by ONNXRuntime.<br>

keras2onnx/_builtin.py

@@ -9,6 +9,7 @@
 import numpy as np
 from typing import Union
 from onnx import numpy_helper, mapping
+from .common.utils import count_dynamic_dim
 from .common.onnx_ops import apply_identity, apply_reshape, OnnxOperatorBuilder
 from .funcbook import converter_func, set_converters
 from .proto import keras
@@ -492,7 +493,7 @@ def convert_tf_depthwise_conv2d(scope, operator, container):
     if node.get_attr('padding') == b'VALID':
         attrs['auto_pad'] = 'VALID'
     elif node.get_attr('padding') == b'SAME':
-        if input_shape.count(None) > 1:
+        if count_dynamic_dim(input_shape) > 1:
             attrs['auto_pad'] = 'SAME_UPPER'
         else:
             attrs['auto_pad'] = 'NOTSET'

keras2onnx/common/utils.py

@@ -54,6 +54,14 @@ def get_default_batch_size():
     return 'N'
 
 
+def count_dynamic_dim(shape):
+    num = 0
+    for s_ in shape:
+        if isinstance(s_, int) and s_ >= 0:
+            num += 1
+    return len(shape) - num
+
+
 def get_producer():
     """
     Internal helper function to return the producer

keras2onnx/ke2onnx/conv.py

@@ -7,6 +7,7 @@
 from .activation import activation_map
 from ..proto import keras
 from ..proto import onnx_proto
+from ..common.utils import count_dynamic_dim
 from ..common.onnx_ops import (apply_identity, apply_pad, apply_softmax,
                                apply_transpose, apply_mul, apply_sigmoid)
 
@@ -144,7 +145,7 @@ def convert_keras_conv_core(scope, operator, container, is_transpose, n_dims, in
     if op.padding == 'valid':
         attrs['auto_pad'] = 'VALID'
     elif op.padding == 'same':
-        if input_shape.count(None) > 1:
+        if count_dynamic_dim(input_shape) > 1:
             if is_transpose:
                 attrs['auto_pad'] = 'SAME_LOWER'  # the controversial def in onnx spec.
             else: