Fix a bug that multiple (conv, batch_norm) ops could not be optimized.

README.md

@@ -17,8 +17,8 @@ The common issues we run into we try to document here [Troubleshooting Guide](Tr
 
 | Build Type | OS | Python | TensorFlow | ONNX opset | Status |
 | ---        | ---    | ---    | ---        | ---        | ---    |
-| Unit Test - Basic | Linux, MacOS<sup>\*</sup>, Windows<sup>\*</sup> | 3.7-3.10 | 1.13-1.15, 2.1-2.11 | 14-18 | [![Build Status](https://dev.azure.com/tensorflow-onnx/tensorflow-onnx/_apis/build/status/unit_test?branchName=main)](https://dev.azure.com/tensorflow-onnx/tensorflow-onnx/_build/latest?definitionId=16&branchName=main) |
-| Unit Test - Full | Linux, MacOS, Windows | 3.7-3.10 | 1.13-1.15, 2.1-2.11 | 14-18 | [![Build Status](https://dev.azure.com/tensorflow-onnx/tensorflow-onnx/_apis/build/status/unit_test-matrix?branchName=main)](https://dev.azure.com/tensorflow-onnx/tensorflow-onnx/_build/latest?definitionId=18&branchName=main) | |
+| Unit Test - Basic | Linux, MacOS<sup>\*</sup>, Windows<sup>\*</sup> | 3.7-3.10 | 1.15, 2.1-2.11 | 14-18 | [![Build Status](https://dev.azure.com/tensorflow-onnx/tensorflow-onnx/_apis/build/status/unit_test?branchName=main)](https://dev.azure.com/tensorflow-onnx/tensorflow-onnx/_build/latest?definitionId=16&branchName=main) |
+| Unit Test - Full | Linux, MacOS, Windows | 3.7-3.10 | 1.15, 2.1-2.11 | 14-18 | [![Build Status](https://dev.azure.com/tensorflow-onnx/tensorflow-onnx/_apis/build/status/unit_test-matrix?branchName=main)](https://dev.azure.com/tensorflow-onnx/tensorflow-onnx/_build/latest?definitionId=18&branchName=main) | |
 <br/>
 
 ## Supported Versions

ci_build/azure_pipelines/onnxruntime_nightly_test.yml

@@ -16,18 +16,6 @@ stages:
           - template: 'unit_test.yml'
         report_coverage: 'True'
 
-    - template: 'templates/job_generator.yml'
-      parameters:
-        platforms: ['linux', 'windows']
-        python_versions: ['3.7']
-        tf_versions: ['1.14.0']
-        onnx_opsets: ['']
-        onnx_backends: {onnxruntime: ['nightly']}
-        job:
-          steps:
-          - template: 'unit_test.yml'
-        report_coverage: 'True'
-
     - template: 'templates/job_generator.yml'
       parameters:
         platforms: ['linux', 'windows']

ci_build/azure_pipelines/pretrained_model_test-matrix.yml

@@ -1,15 +1,6 @@
 # Pre-trained model test, full matrix
 
 jobs:
-- template: 'templates/job_generator.yml'
-  parameters:
-    platforms: ['linux', 'windows']
-    python_versions: ['3.7']
-    tf_versions: ['1.14.0']
-    job:
-      steps:
-      - template: 'pretrained_model_test.yml'
-
 - template: 'templates/job_generator.yml'
   parameters:
     platforms: ['linux', 'windows']

ci_build/azure_pipelines/unit_test-matrix.yml

@@ -7,7 +7,7 @@ stages:
       parameters:
         platforms: ['linux', 'windows']
         python_versions: ['3.7']
-        tf_versions: ['1.14.0', '1.15.2']
+        tf_versions: ['1.15.2']
         onnx_opsets: ['']
         job:
           steps:

ci_build/azure_pipelines/unit_test.yml

@@ -122,16 +122,6 @@ stages:
           - template: 'unit_test.yml'
         report_coverage: 'True'
 
-    - template: 'templates/job_generator.yml'
-      parameters:
-        platforms: ['windows']
-        tf_versions: ['1.14.0']
-        onnx_opsets: ['14']
-        job:
-          steps:
-          - template: 'unit_test.yml'
-        report_coverage: 'True'
-
     - template: 'templates/job_generator.yml'
       parameters:
         python_versions: ['3.8']

tests/test_backend.py

@@ -3087,6 +3087,57 @@ def graph_validator(g):
 
         self._run_test_case(func_fusedbn, [_OUTPUT], {_INPUT: x_val}, rtol=1e-05, graph_validator=graph_validator)
 
+    @check_opset_min_version(7, "batchnorm")
+    def test_multiple_conv2d_fused_batchnorm(self):
+        x_shape = [1, 28, 28, 2]
+        x_val = np.random.random_sample(x_shape).astype(np.float32)
+        w = np.array([[2., 1., 1.],
+                      [1., 3., 1.],
+                      [1., 1., 4.]], dtype=np.float32).reshape(_KERNEL3x3)
+        # 2 channels for input and output
+        w = np.concatenate([w, w, w, w]).reshape([3, 3, 2, 2])
+        scale_dtype = np.float32
+        scale_shape = x_shape[-1:]
+        scale_val = np.random.random_sample(scale_shape).astype(scale_dtype)
+        offset_val = np.random.random_sample(scale_shape).astype(scale_dtype)
+        mean_val = np.random.random_sample(scale_shape).astype(scale_dtype)
+        var_val = np.random.random_sample(scale_shape).astype(scale_dtype)
+
+        def func_conv2d(x):
+            kernel = tf.constant(w, dtype=tf.float32, name='k')
+            conv = tf.nn.conv2d(x, kernel, strides=[1, 1, 1, 1], padding='VALID')
+            return conv
+
+        def func_multiple_fusedbn(x):
+            scale = tf.constant(scale_val, name='scale')
+            offset = tf.constant(offset_val, name='offset')
+            mean = tf.constant(mean_val, name='mean')
+            var = tf.constant(var_val, name='variance')
+            epsilon = 0.1234
+            y, _, _ = fused_batch_norm(
+                func_conv2d(x), scale, offset, mean=mean, variance=var,
+                epsilon=epsilon, data_format='NHWC', is_training=False)
+
+            y = tf.nn.relu(y)
+
+            y, _, _ = fused_batch_norm(
+                func_conv2d(y), scale, offset, mean=mean, variance=var,
+                epsilon=epsilon, data_format='NHWC', is_training=False)
+
+            y, _, _ = fused_batch_norm(
+                func_conv2d(y), scale, offset, mean=mean, variance=var,
+                epsilon=epsilon, data_format='NHWC', is_training=False)
+
+            return tf.identity(y, name=_TFOUTPUT)
+
+        def graph_validator(g):
+            if 'BatchNormalization' in [n.type for n in g.get_nodes()]:
+                return False
+            return True
+
+        self._run_test_case(func_multiple_fusedbn, [_OUTPUT], {_INPUT: x_val}, rtol=1e-05,
+                            graph_validator=graph_validator)
+
     @check_tf_min_version("1.15")
     @check_opset_min_version(10, "quantize_and_dequantize")
     def test_qdq_unsigned_input(self):

tf2onnx/optimizer/back_to_back_optimizer.py

@@ -41,9 +41,10 @@ def _optimize_at_current_graph_level(self, g):
 
             # topological sort of candidates
             # simplifying assumption for back-to-back-optimizer is
-            # the op_types have 1 input, 1 output, but multiple consumers
+            # the op_types have 1 input, 1 output, but multiple consumers.
+            # if optype contains 2 elements, the second element should not be considered as a consumer.
             has_dependencies = set()
-            consumer_node_ids = {n.output[0]: [] for n in nodes}
+            consumer_node_ids = {n.output[0]: [] for n in nodes if len(optype) < 2 or n.type == optype[0]}
             for n in nodes:
                 if n.input[0] in consumer_node_ids:
                     consumer_node_ids[n.input[0]].extend([n])