Merge branch 'transform' into 'master'

Refactor caffe depthwise

See merge request !481

mace/python/tools/converter_tool/caffe_converter.py

@@ -144,6 +144,7 @@ class CaffeConverter(base_converter.ConverterInterface):
         'ReLU': ActivationType.RELU,
         'PReLU': ActivationType.PRELU,
         'TanH': ActivationType.TANH,
+        'Sigmoid': ActivationType.SIGMOID,
     }
 
     def __init__(self, option, src_model_file, src_weight_file):
@@ -337,10 +338,15 @@ class CaffeConverter(base_converter.ConverterInterface):
         param = caffe_op.layer.convolution_param
         is_depthwise = False
         if param.HasField(caffe_group_str):
-            mace_check(param.group == caffe_op.blob[0].shape[1] and
-                       caffe_op.blob[0].shape[0] == 1,
+            filter_data = caffe_op.blobs[0]
+            mace_check(param.group == filter_data.shape[0] and
+                       filter_data.shape[1] == 1,
                        "Mace do not support group convolution yet")
             is_depthwise = True
+            caffe_op.blobs[0] = filter_data.reshape(1,
+                                                    filter_data.shape[0],
+                                                    filter_data.shape[2],
+                                                    filter_data.shape[3])
 
         if is_depthwise:
             op.type = MaceOp.DepthwiseConv2d.name

mace/python/tools/converter_tool/shape_inference.py

@@ -17,6 +17,7 @@ class ShapeInference(object):
     def __init__(self, net, input_nodes):
         self._op_shape_inference = {
             MaceOp.Conv2D.name: self.infer_shape_conv_pool_shape,
+            MaceOp.DepthwiseConv2d.name: self.infer_shape_conv_pool_shape,
             MaceOp.Eltwise.name: self.infer_shape_general,
             MaceOp.FoldedBatchNorm.name: self.infer_shape_general,
             MaceOp.AddN.name: self.infer_shape_general,
@@ -104,7 +105,10 @@ class ShapeInference(object):
         if ConverterUtil.data_format(op) == DataFormat.NCHW \
                 and ConverterUtil.filter_format(self._net) == FilterFormat.OIHW:  # noqa
             # filter format: OIHW
-            output_shape[1] = filter_shape[0]
+            if op.type == MaceOp.DepthwiseConv2d.name:
+                output_shape[1] = filter_shape[0] * filter_shape[1]
+            else:
+                output_shape[1] = filter_shape[0]
             output_shape[2] = int(
                 round_func((input_shape[2] + paddings[0] - filter_shape[2] -
                             (filter_shape[2] - 1) *

mace/python/tools/converter_tool/transformer.py

@@ -692,7 +692,7 @@ class Transformer(base_converter.ConverterInterface):
                             filter_data = filter_data.transpose(2, 3, 0, 1)
                             filter.float_data[:] = filter_data.flat
                             filter.dims[:] = filter_data.shape
-                        elif op.type == MaceOp.Depthwiseconv2d.name:
+                        elif op.type == MaceOp.DepthwiseConv2d.name:
                             filter_data = filter_data.transpose(2, 3, 1, 0)
                             filter.float_data[:] = filter_data.flat
                             filter.dims[:] = filter_data.shape