提交 94b1bae6 编写于 作者: W weishengyu

dbg

上级 1153ad65
......@@ -503,7 +503,9 @@ class HRNet(TheseusLayer):
self.tr1_2 = ConvBNLayer(
num_channels=256,
num_filters=width * 2,
filter_size=3)
filter_size=3,
stride=2
)
self.st2 = Stage(
num_channels=channels_2,
......@@ -515,7 +517,9 @@ class HRNet(TheseusLayer):
self.tr2 = ConvBNLayer(
num_channels=width * 2,
num_filters=width * 4,
filter_size=3)
filter_size=3,
stride=2
)
self.st3 = Stage(
num_channels=channels_3,
num_modules=num_modules_3,
......@@ -526,7 +530,9 @@ class HRNet(TheseusLayer):
self.tr3 = ConvBNLayer(
num_channels=width * 4,
num_filters=width * 8,
filter_size=3)
filter_size=3,
stride=2
)
self.st4 = Stage(
num_channels=channels_4,
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册