From 8e26493c329f07623438c6c32824bb7bc8e29aa9 Mon Sep 17 00:00:00 2001
From: littletomatodonkey <2120160898@bit.edu.cn>
Date: Thu, 7 Jan 2021 19:24:29 +0800
Subject: [PATCH] fix lr mult val (#551)

---
 docs/en/tutorials/quick_start_en.md       |  2 +-
 docs/zh_CN/tutorials/quick_start.md       |  2 +-
 ppcls/modeling/architectures/resnet_vd.py | 10 ++++++++--
 3 files changed, 10 insertions(+), 4 deletions(-)

diff --git a/docs/en/tutorials/quick_start_en.md b/docs/en/tutorials/quick_start_en.md
index 473c43ef..840ce547 100644
--- a/docs/en/tutorials/quick_start_en.md
+++ b/docs/en/tutorials/quick_start_en.md
@@ -107,7 +107,7 @@ Note: when finetuning model, which has been trained by SSLD, please use smaller
 ARCHITECTURE:
     name: 'ResNet50_vd'
     params:
-        lr_mult_list: [0.1, 0.1, 0.2, 0.2, 0.3]
+        lr_mult_list: [0.5, 0.5, 0.6, 0.6, 0.8]
 pretrained_model: "./pretrained/ResNet50_vd_ssld_pretrained"
 ```
 
diff --git a/docs/zh_CN/tutorials/quick_start.md b/docs/zh_CN/tutorials/quick_start.md
index 5c88bd0c..09c35a9d 100644
--- a/docs/zh_CN/tutorials/quick_start.md
+++ b/docs/zh_CN/tutorials/quick_start.md
@@ -108,7 +108,7 @@ python -m paddle.distributed.launch \
 ARCHITECTURE:
     name: 'ResNet50_vd'
     params:
-        lr_mult_list: [0.1, 0.1, 0.2, 0.2, 0.3]
+        lr_mult_list: [0.5, 0.5, 0.6, 0.6, 0.8]
 pretrained_model: "./pretrained/ResNet50_vd_ssld_pretrained"
 ```
 
diff --git a/ppcls/modeling/architectures/resnet_vd.py b/ppcls/modeling/architectures/resnet_vd.py
index 8b725477..5867261d 100644
--- a/ppcls/modeling/architectures/resnet_vd.py
+++ b/ppcls/modeling/architectures/resnet_vd.py
@@ -64,8 +64,10 @@ class ConvBNLayer(nn.Layer):
         self._batch_norm = BatchNorm(
             num_filters,
             act=act,
-            param_attr=ParamAttr(name=bn_name + '_scale'),
-            bias_attr=ParamAttr(bn_name + '_offset'),
+            param_attr=ParamAttr(
+                name=bn_name + '_scale', learning_rate=lr_mult),
+            bias_attr=ParamAttr(
+                bn_name + '_offset', learning_rate=lr_mult),
             moving_mean_name=bn_name + '_mean',
             moving_variance_name=bn_name + '_variance')
 
@@ -118,6 +120,7 @@ class BottleneckBlock(nn.Layer):
                 filter_size=1,
                 stride=1,
                 is_vd_mode=False if if_first else True,
+                lr_mult=lr_mult,
                 name=name + "_branch1")
 
         self.shortcut = shortcut
@@ -153,12 +156,14 @@ class BasicBlock(nn.Layer):
             filter_size=3,
             stride=stride,
             act='relu',
+            lr_mult=lr_mult,
             name=name + "_branch2a")
         self.conv1 = ConvBNLayer(
             num_channels=num_filters,
             num_filters=num_filters,
             filter_size=3,
             act=None,
+            lr_mult=lr_mult,
             name=name + "_branch2b")
 
         if not shortcut:
@@ -168,6 +173,7 @@ class BasicBlock(nn.Layer):
                 filter_size=1,
                 stride=1,
                 is_vd_mode=False if if_first else True,
+                lr_mult=lr_mult,
                 name=name + "_branch1")
 
         self.shortcut = shortcut
-- 
GitLab