diff --git a/src/operators/kernel/central-arm-func/pool_arm_func.h b/src/operators/kernel/central-arm-func/pool_arm_func.h
index 37479c22efe95b6506054cf3ded5855aa766c34c..1c22a2646ea4efd91e6c73333c21e0d0c56fcb17 100644
--- a/src/operators/kernel/central-arm-func/pool_arm_func.h
+++ b/src/operators/kernel/central-arm-func/pool_arm_func.h
@@ -76,7 +76,7 @@ void PoolCompute(const PoolParam<CPU> &param) {
       }
     }
 
-  } else if (ksize[0] == 2 && ksize[0] == ksize[1] && strides[0] == 2 &&
+  } else if (0 && ksize[0] == 2 && ksize[0] == ksize[1] && strides[0] == 2 &&
              strides[0] == strides[1] && paddings[0] == paddings[1] &&
              paddings[1] == 0) {
 #if __ARM_NEON
diff --git a/src/operators/math/im2col.cpp b/src/operators/math/im2col.cpp
index 090ccdf24e214fc86b8a4032df228d50caa65ef9..e23b02486bec81c4685420d451dc093dd420ac97 100644
--- a/src/operators/math/im2col.cpp
+++ b/src/operators/math/im2col.cpp
@@ -78,7 +78,7 @@ class Im2ColFunctor<ColFormat::kCFO, CPU, T> {
          (((isize - 2 * padding[0] + filter_height) % stride[0] == 0) ? 1 : 0));
     int fill = isize % 2;
     if (stride[0] == 1 && filter_height == 3 && pad1 && pad2 &&
-        dilation[0] == 1 && im_height > 2) {
+        dilation[0] == 1 && im_height > 2 && im_height == im_width) {
       for (int c = 0; c < im_channels; ++c) {
         int oosize = osize * osize;
         int nk4 = osize / 4;
@@ -250,7 +250,7 @@ class Im2ColFunctor<ColFormat::kCFO, CPU, T> {
         im_data += isize * isize;
       }
     } else if (stride[0] == 2 && filter_height == 3 && pad1 &&
-               dilation[0] == 1 && im_height > 2) {
+               dilation[0] == 1 && im_height > 2 && im_height == im_width) {
       for (int c = 0; c < im_channels; ++c) {
         int oosize = osize * osize;
         int nk4 = osize / 4;
@@ -528,7 +528,6 @@ class Im2ColFunctor<ColFormat::kOCF, CPU, T> {
     int filter_width = col->dims()[4];
     int col_height = col->dims()[0];
     int col_width = col->dims()[1];
-
     //    PADDLE_ENFORCE_EQ(
     //        (im_height + padding[0] + padding[2] -
     //        filter_height) / stride[0]
@@ -544,7 +543,6 @@ class Im2ColFunctor<ColFormat::kOCF, CPU, T> {
 
     const T *im_data = im.data<T>();
     T *col_data = col->data<T>();
-
     for (int col_row_idx = 0; col_row_idx < col_height; ++col_row_idx) {
       for (int col_col_idx = 0; col_col_idx < col_width; ++col_col_idx) {
         for (int channel = 0; channel < im_channels; ++channel) {
@@ -556,7 +554,6 @@ class Im2ColFunctor<ColFormat::kOCF, CPU, T> {
                  ++filter_col_idx) {
               int im_col_offset =
                   col_col_idx * stride[1] + filter_col_idx - padding[1];
-
               int col_offset =
                   ((((col_row_idx)*col_width + col_col_idx) * im_channels +
                     channel) *
@@ -564,7 +561,6 @@ class Im2ColFunctor<ColFormat::kOCF, CPU, T> {
                    filter_row_idx) *
                       filter_width +
                   filter_col_idx;
-
               int im_offset = (channel * im_height + im_row_offset) * im_width +
                               im_col_offset;
               col_data[col_offset] =
diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt
index 1893491f1163c42784af1213b6581ae7817f86b2..e52db8b8f04bd9e4ae86241542ea2565937c677b 100644
--- a/test/CMakeLists.txt
+++ b/test/CMakeLists.txt
@@ -339,6 +339,10 @@ if (NOT FOUND_MATCH)
     ADD_EXECUTABLE(test-multi-process net/test_multi_inference_predict.cpp test_helper.h test_include.h)
     target_link_libraries(test-multi-process paddle-mobile)
 
+    # gen test
+    ADD_EXECUTABLE(test-eng net/test_eng.cpp test_helper.h test_include.h)
+    target_link_libraries(test-eng paddle-mobile)
+
 
     #add_library(test-lib-size SHARED common/test_lib_size.h common/test_lib_size.cpp)
 endif ()
diff --git a/test/net/test_eng.cpp b/test/net/test_eng.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..d10cb33a6922f4a60b798211ab4415312b86796c
--- /dev/null
+++ b/test/net/test_eng.cpp
@@ -0,0 +1,50 @@
+/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License. */
+
+#include <iostream>
+#include "../test_helper.h"
+#include "../test_include.h"
+
+int main() {
+#ifdef PADDLE_MOBILE_CPU
+  paddle_mobile::PaddleMobile<paddle_mobile::CPU> paddle_mobile;
+#endif
+  //    paddle_mobile.SetThreadNum(4);
+  auto time1 = time();
+  if (paddle_mobile.Load(std::string(g_eng) + "/model",
+                         std::string(g_eng) + "/params", false, false, 1,
+                         true)) {
+    auto time2 = time();
+    std::cout << "load cost :" << time_diff(time1, time1) << "ms" << std::endl;
+    std::vector<int64_t> dims{1, 1, 48, 48};
+    LoDTensor input_tensor;
+    SetupTensor<float>(&input_tensor, {1, 1, 48, 48}, static_cast<float>(0),
+                       static_cast<float>(1));
+
+    std::vector<float> input(input_tensor.data<float>(),
+                             input_tensor.data<float>() + input_tensor.numel());
+    //   预热十次
+    for (int i = 0; i < 1; ++i) {
+      paddle_mobile.PredictLod(input_tensor);
+    }
+    auto time3 = time();
+    //        for (int i = 0; i < 10; ++i) {
+    //            paddle_mobile.Predict(input, dims);
+    //        }
+    auto time4 = time();
+    std::cout << "predict cost :" << time_diff(time3, time4) << "ms"
+              << std::endl;
+  }
+  return 0;
+}