update number of channels after transition_block (d2l-ai#525)

panhongjin · Apr 25, 2019 · 61728ff · 61728ff
1 parent ca4a41a
commit 61728ff
Showing 1 changed file with 2 additions and 1 deletion.
diff --git a/chapter_convolutional-neural-networks/densenet.md b/chapter_convolutional-neural-networks/densenet.md
@@ -98,7 +98,8 @@ for i, num_convs in enumerate(num_convs_in_dense_blocks):
     num_channels += num_convs * growth_rate
     # 在稠密块之间加入通道数减半的过渡层
     if i != len(num_convs_in_dense_blocks) - 1:
-        net.add(transition_block(num_channels // 2))
+        num_channels //= 2
+        net.add(transition_block(num_channels))
 ```
 
 同ResNet一样，最后接上全局池化层和全连接层来输出。