Added support for converting NonMaxSuppression in opset 10 with dynamic padding

Signed-off-by: Tom Wildenhain <[email protected]>

Author: TomWildenhain-Microsoft

tests/test_backend.py

@@ -2822,7 +2822,7 @@ def func(boxes, scores):
 
         self._run_test_case(func, [_OUTPUT, _OUTPUT1], {_INPUT: boxes_val, _INPUT1: scores_val})
 
-    @check_opset_min_version(11, "NonMaxSuppressionV4")
+    @check_opset_min_version(10, "NonMaxSuppressionV4")
     def test_non_max_suppression_v4(self):
         box_num = 10
         boxes_val = np.random.random_sample([box_num, 4]).astype(np.float32)
@@ -2835,7 +2835,7 @@ def func(boxes, scores):
 
         self._run_test_case(func, [_OUTPUT, _OUTPUT1], {_INPUT: boxes_val, _INPUT1: scores_val})
 
-    @check_opset_min_version(11, "NonMaxSuppressionV4")
+    @check_opset_min_version(10, "NonMaxSuppressionV4")
     def test_non_max_suppression_v4_no_padding(self):
         box_num = 10
         boxes_val = np.random.random_sample([box_num, 4]).astype(np.float32)
@@ -2849,7 +2849,7 @@ def func(boxes, scores):
         self._run_test_case(func, [_OUTPUT, _OUTPUT1], {_INPUT: boxes_val, _INPUT1: scores_val})
 
     @check_tf_min_version("1.15")
-    @check_opset_min_version(11, "NonMaxSuppressionV5")
+    @check_opset_min_version(10, "NonMaxSuppressionV5")
     def test_non_max_suppression_v5(self):
         box_num = 10
         boxes_val = np.random.random_sample([box_num, 4]).astype(np.float32)

tf2onnx/onnx_opset/tensor.py

@@ -13,7 +13,7 @@
 import sys
 
 import numpy as np
-from onnx import onnx_pb
+from onnx import onnx_pb, helper
 from onnx.onnx_pb import TensorProto
 
 from tf2onnx import constants, utils
@@ -1485,13 +1485,18 @@ def version_10(cls, ctx, node, **kwargs):
                           outputs=[node.output[1]], dtypes=dtypes[1], shapes=shapes[1])
         elif "pad_to_max_output_size" in node.attr:  # V4
             shape_op = ctx.make_node("Shape", inputs=[squeeze_op.output[0]])
-            const_zero = ctx.make_const(utils.make_name("const_zero"), np.array([0], dtype=np.int64))
             sub_op = ctx.make_node("Sub", inputs=[max_output_size, shape_op.output[0]])
-            raw_pad = ctx.make_node("Concat", inputs=[const_zero.output[0], sub_op.output[0]], attr={'axis': 0})
-            raw_pad_float = ctx.make_node("Cast", inputs=[raw_pad.output[0]], attr={"to": onnx_pb.TensorProto.FLOAT})
+            raw_pad_float = ctx.make_node("Cast", inputs=[sub_op.output[0]], attr={"to": onnx_pb.TensorProto.FLOAT})
             relu_op = ctx.make_node("Relu", inputs=[raw_pad_float.output[0]])
-            pad_val = ctx.make_node("Cast", inputs=[relu_op.output[0]], attr={"to": onnx_pb.TensorProto.INT64})
-            pad_op = ctx.make_node("Pad", inputs=[squeeze_op.output[0], pad_val.output[0]])
+            pad_amt = ctx.make_node("Cast", inputs=[relu_op.output[0]], attr={"to": onnx_pb.TensorProto.INT64})
+            if ctx.opset <= 10:  # Dynamic padding not supported before opset 11
+                zero_tensor = helper.make_tensor("value", onnx_pb.TensorProto.INT64, dims=[1], vals=[0])
+                padding = ctx.make_node("ConstantOfShape", inputs=[pad_amt.output[0]], attr={"value": zero_tensor})
+                pad_op = ctx.make_node("Concat", inputs=[squeeze_op.output[0], padding.output[0]], attr={'axis': 0})
+            else:
+                const_zero = ctx.make_const(utils.make_name("const_zero"), np.array([0], dtype=np.int64))
+                pad_val = ctx.make_node("Concat", inputs=[const_zero.output[0], pad_amt.output[0]], attr={'axis': 0})
+                pad_op = ctx.make_node("Pad", inputs=[squeeze_op.output[0], pad_val.output[0]])
             ctx.make_node("Cast", inputs=pad_op.output, name="cast_A", attr={"to": onnx_pb.TensorProto.INT32},
                           outputs=[node.output[0]], dtypes=dtypes[0], shapes=shapes[0], op_name_scope=node.name)
             reduce_op = ctx.make_node("ReduceSum", inputs=shape_op.output, attr={"axes": [0], "keepdims": 0})