feat: Implementated average poolin 2d.

The supported args cannot be tuple right now. This is WIP. Will extend
it to support tuple arguments.

Addresses: issue pfnet-research#160

elichika/elichika/chainer2onnx.py

@@ -44,6 +44,7 @@ def compile_model(model, inputs) -> 'ONNXModel':
     oc.chainer_f_converter[F.softmax] = fb.convert_softmax
     oc.chainer_f_converter[F.pad_sequence] = fb.convert_pad_sequence
     oc.chainer_f_converter[F.softmax_cross_entropy] = fb.convert_softmax_cross_entropy
+    oc.chainer_f_converter[F.average_pooling_2d] = fb.convert_average_pool_2d
 
     # assign names
     oc.assigned_names.clear()

elichika/elichika/functions_buildin.py

@@ -66,3 +66,36 @@ def convert_softmax_cross_entropy(onnx_graph, node):
         node.inputs,
         node.outputs,
         str(node.lineprop))
+
+
+def convert_average_pool_2d(onnx_graph, node):
+    def _pair(x):
+        if isinstance(x, collections.Iterable):
+            return x
+        return (x, x)
+
+    kwargs = {}
+    ksize = oc.try_get_attribute(node.inputs[1])
+    kwargs['kernel_shape'] = _pair(ksize)
+
+    value = oc.try_get_attribute(node.inputs[2])
+    if value is not None:
+        kwargs['strides'] = _pair(value)
+    else:
+        kwargs['strides'] = _pair(ksize)
+
+    value = oc.try_get_attribute(node.inputs[3])
+    if value is not None:
+        kwargs['pads'] = _pair(value) * 2
+    else:
+        kwargs['pads'] = _pair(0)
+
+    kwargs['count_include_pad'] = 1
+
+    onnx_graph.add_node(
+        "AveragePool",
+        [node.inputs[0]],
+        [node.outputs[0]],
+        name=str(node.lineprop),
+        **kwargs,
+        )

elichika/elichika/parser/core.py

@@ -63,6 +63,9 @@ def instance_converter(m, i):
             functions_builtin.ChainerFunction(F.pad_sequence), None)
         f_dict.get_field().get_attribute('pad_sequence').revise(
             values.ValueRef(f_pad_sequence))
+        f_average_pooling_2d = values.FuncValue(
+            functions_builtin.ChainerFunction(F.average_pooling_2d), None)
+        f_dict.get_field().get_attribute('average_pooling_2d').revise(values.ValueRef(f_average_pooling_2d))
         default_module.set_default_value(chainer_functions_module_name, f_dict)
 
     # numpy

elichika/tests/node/AveragePool2d.py

@@ -0,0 +1,53 @@
+# coding: utf-8
+
+import chainer
+import chainer.functions as F
+
+
+class AvgPool(chainer.Chain):
+
+    def __init__(self):
+        super(AvgPool, self).__init__()
+
+    def forward(self, x):
+        y1 = F.average_pooling_2d(x, 1, stride=2)
+        return y1
+
+
+class AvgPoolPad(chainer.Chain):
+
+    def __init__(self):
+        super(AvgPoolPad, self).__init__()
+
+    def forward(self, x):
+        y1 = F.average_pooling_2d(x, 3, stride=1, pad=2)
+        return y1
+
+
+class AvgPoolNoStride(chainer.Chain):
+
+    def __init__(self):
+        super(AvgPoolNoStride, self).__init__()
+
+    def forward(self, x):
+        y1 = F.average_pooling_2d(x, 3)
+        return y1
+
+
+# ======================================
+
+import testtools
+import numpy as np
+
+
+def main():
+    np.random.seed(123)
+    x = np.random.rand(2, 20, 15, 17).astype(np.float32)
+
+    testtools.generate_testcase(AvgPool(), [x], subname='default')
+    testtools.generate_testcase(AvgPoolPad(), [x], subname='withpad')
+    testtools.generate_testcase(AvgPoolNoStride(), [x], subname='withoutstride')
+
+
+if __name__ == '__main__':
+    main()