Merge branch 'fix_nas' into 'develop'

update mobilenetv2 space

See merge request !50

paddleslim/nas/search_space/mobilenetv2.py

@@ -145,7 +145,18 @@ class MobileNetV2Space(SearchSpaceBase):
                 (self.multiply[tokens[25]], self.filter_num6[tokens[26]],
                  self.repeat[tokens[27]], 1, self.k_size[tokens[28]]))
 
-        def net_arch(input):
+        def net_arch(input, end_points=None, decode_points=None):
+            decode_ends = dict()
+
+            def check_points(count, points):
+                if points is None:
+                    return False
+                else:
+                    if isinstance(points, list):
+                        return (True if count in points else False)
+                    else:
+                        return (True if count == points else False)
+
             #conv1
             # all padding is 'SAME' in the conv2d, can compute the actual padding automatic. 
             input = conv_bn_layer(
@@ -156,6 +167,12 @@ class MobileNetV2Space(SearchSpaceBase):
                 padding='SAME',
                 act='relu6',
                 name='mobilenetv2_conv1_1')
+            layer_count = 1
+            if check_points(layer_count, decode_points):
+                decode_ends[layer_count] = input
+
+            if check_points(layer_count, end_points):
+                return input, decode_ends
 
             # bottleneck sequences
             i = 1
@@ -173,6 +190,13 @@ class MobileNetV2Space(SearchSpaceBase):
                     k=k,
                     name='mobilenetv2_conv' + str(i))
                 in_c = int(c * self.scale)
+                layer_count += n
+
+                if check_points(layer_count, decode_points):
+                    decode_ends[layer_count] = depthwise_output
+
+                if check_points(layer_count, end_points):
+                    return input, decode_ends
 
             # last conv
             input = conv_bn_layer(
@@ -266,6 +290,8 @@ class MobileNetV2Space(SearchSpaceBase):
             name=name + '_dwise',
             use_cudnn=False)
 
+        depthwise_output = bottleneck_conv
+
         linear_out = conv_bn_layer(
             input=bottleneck_conv,
             num_filters=num_filters,
@@ -278,7 +304,7 @@ class MobileNetV2Space(SearchSpaceBase):
         out = linear_out
         if ifshortcut:
             out = self._shortcut(input=input, data_residual=out)
-        return out
+        return out, depthwise_output
 
     def _invresi_blocks(self, input, in_c, t, c, n, s, k, name=None):
         """Build inverted residual blocks.
@@ -308,7 +334,7 @@ class MobileNetV2Space(SearchSpaceBase):
         last_c = c
 
         for i in range(1, n):
-            last_residual_block = self._inverted_residual_unit(
+            last_residual_block, depthwise_output = self._inverted_residual_unit(
                 input=last_residual_block,
                 num_in_filter=last_c,
                 num_filters=c,
@@ -317,4 +343,4 @@ class MobileNetV2Space(SearchSpaceBase):
                 filter_size=k,
                 expansion_factor=t,
                 name=name + '_' + str(i + 1))
-        return last_residual_block
+        return last_residual_block, depthwise_output

paddleslim/nas/search_space/search_space_base.py

@@ -19,16 +19,19 @@ class SearchSpaceBase(object):
     """Controller for Neural Architecture Search.
     """
 
-    def __init__(self,
-                 input_size,
-                 output_size,
-                 block_num,
-                 block_mask=None,
-                 *argss):
+    def __init__(self, input_size, output_size, block_num, block_mask, *args):
+        """init model config
+        """
         self.input_size = input_size
         self.output_size = output_size
         self.block_num = block_num
         self.block_mask = block_mask
+        if self.block_mask is not None:
+            assert isinstance(self.block_mask,
+                              list), 'Block_mask must be a list.'
+            print(
+                "If block_mask is NOT None, we will use block_mask as major configs!"
+            )
 
     def init_tokens(self):
         """Get init tokens in search space.