diff --git a/PaddleCV/image_classification/README.md b/PaddleCV/image_classification/README.md
index bf4f509d67a05adb0d04298bd202b72612c572cb..7b469d1b982cdb6cfc59edfd07f85c4b01641eb3 100644
--- a/PaddleCV/image_classification/README.md
+++ b/PaddleCV/image_classification/README.md
@@ -64,6 +64,7 @@ python train.py \
        --image_shape=3,224,224 \
        --model_save_dir=output/ \
        --with_mem_opt=False \
+       --with_inplace=True \
        --lr_strategy=piecewise_decay \
        --lr=0.1
 ```
@@ -76,7 +77,8 @@ python train.py \
 * **class_dim**: the class number of the classification task. Default: 1000.
 * **image_shape**: input size of the network. Default: "3,224,224".
 * **model_save_dir**: the directory to save trained model. Default: "output".
-* **with_mem_opt**: whether to use memory optimization or not. Default: True.
+* **with_mem_opt**: whether to use memory optimization or not. Default: False.
+* **with_inplace**: whether to use inplace memory optimization or not. Default: True.
 * **lr_strategy**: learning rate changing strategy. Default: "piecewise_decay".
 * **lr**: initialized learning rate. Default: 0.1.
 * **pretrained_model**: model path for pretraining. Default: None.
@@ -185,4 +187,3 @@ Available top-1/top-5 validation accuracy on ImageNet 2012 are listed in table.
 |[GoogleNet](https://paddle-imagenet-models-name.bj.bcebos.com/GoogleNet_pretrained.tar) | 70.70%/89.66% |
 |[ShuffleNetV2](https://paddle-imagenet-models-name.bj.bcebos.com/ShuffleNetV2_pretrained.tar) | 70.03%/89.17% |
 |[InceptionV4](https://paddle-imagenet-models-name.bj.bcebos.com/InceptionV4_pretrained.tar) | 80.77%/95.26% |
-
diff --git a/PaddleCV/image_classification/README_cn.md b/PaddleCV/image_classification/README_cn.md
index 4af494e7b961505959ebe3c11e02999ad239145f..94781e5d6da1aee78fa6b960be5e5c78ade5c48e 100644
--- a/PaddleCV/image_classification/README_cn.md
+++ b/PaddleCV/image_classification/README_cn.md
@@ -61,6 +61,7 @@ python train.py \
        --image_shape=3,224,224 \
        --model_save_dir=output/ \
        --with_mem_opt=False \
+       --with_inplace=True \
        --lr_strategy=piecewise_decay \
        --lr=0.1
 ```
@@ -74,6 +75,7 @@ python train.py \
 * **image_shape**: 图片大小，默认值: "3,224,224"
 * **model_save_dir**: 模型存储路径，默认值: "output/"
 * **with_mem_opt**: 是否开启显存优化，默认值: False
+* **with_inplace**: 是否开启inplace显存优化，默认值: True
 * **lr_strategy**: 学习率变化策略，默认值: "piecewise_decay"
 * **lr**: 初始学习率，默认值: 0.1
 * **pretrained_model**: 预训练模型路径，默认值: None
diff --git a/PaddleCV/image_classification/train.py b/PaddleCV/image_classification/train.py
index e72af3d9a0e2699d494d4e6640d338af49454b71..a50b1253b701fb19947f4a391c48cc3d99886696 100644
--- a/PaddleCV/image_classification/train.py
+++ b/PaddleCV/image_classification/train.py
@@ -23,6 +23,7 @@ import sys
 import functools
 import math
 
+
 def set_paddle_flags(flags):
     for key, value in flags.items():
         if os.environ.get(key, None) is None:
@@ -62,7 +63,7 @@ add_arg('num_epochs',       int,   120,                  "number of epochs.")
 add_arg('class_dim',        int,   1000,                 "Class number.")
 add_arg('image_shape',      str,   "3,224,224",          "input image size")
 add_arg('model_save_dir',   str,   "output",             "model save directory")
-add_arg('with_mem_opt',     bool,  True,                 "Whether to use memory optimization or not.")
+add_arg('with_mem_opt',     bool,  False,                 "Whether to use memory optimization or not.")
 add_arg('with_inplace',     bool,  True,                 "Whether to use inplace memory optimization.")
 add_arg('pretrained_model', str,   None,                 "Whether to use pretrained model.")
 add_arg('checkpoint',       str,   None,                 "Whether to resume checkpoint.")
@@ -248,7 +249,7 @@ def net_config(image, model, args, is_train, label=0, y_a=0, y_b=0, lam=0.0):
                     return avg_cost
                 else:
                     cost = calc_loss(epsilon,label,class_dim,softmax_out,use_label_smoothing)
-                    
+
             else:
                 cost = fluid.layers.cross_entropy(input=softmax_out, label=label)
         else:
@@ -256,7 +257,7 @@ def net_config(image, model, args, is_train, label=0, y_a=0, y_b=0, lam=0.0):
             softmax_out1, softmax_out = fluid.layers.softmax(out1), fluid.layers.softmax(out2)
             smooth_out1 = fluid.layers.label_smooth(label=softmax_out1, epsilon=0.0, dtype="float32")
             cost = fluid.layers.cross_entropy(input=softmax_out, label=smooth_out1, soft_label=True)
-        
+
         avg_cost = fluid.layers.mean(cost)
         if args.scale_loss > 1:
             avg_cost = fluid.layers.mean(x=cost) * float(args.scale_loss)
@@ -428,7 +429,8 @@ def train(args):
     use_ngraph = os.getenv('FLAGS_use_ngraph')
     if not use_ngraph:
         build_strategy = fluid.BuildStrategy()
-        build_strategy.memory_optimize = args.with_mem_opt
+        # memopt may affect GC results
+        #build_strategy.memory_optimize = args.with_mem_opt
         build_strategy.enable_inplace = args.with_inplace
         #build_strategy.fuse_all_reduce_ops=1