diff --git a/research/cv/ibnnet/README_CN.md b/research/cv/ibnnet/README_CN.md
index 307fd6725e8896e203057f1f97adcd786f87add4..ae8768d7ed61fabe11c7a043488fa04233ba7592 100644
--- a/research/cv/ibnnet/README_CN.md
+++ b/research/cv/ibnnet/README_CN.md
@@ -15,13 +15,14 @@
             - [分布式训练](#分布式训练)
         - [评估过程](#评估过程)
             - [评估](#评估)
+        - [导出mindir模型](#导出mindir模型)
+        - [推理过程](#推理过程)
+            - [用法](#用法)
+            - [结果](#结果)
 - [模型描述](#模型描述)
     - [性能](#性能)
+        - [训练性能](#训练性能)
         - [评估性能](#评估性能)
-        - [推理性能](#推理性能)
-        - [使用方法](#使用方法)
-            - [推理](#推理)
-            - [迁移学习](#迁移学习)
 - [随机情况说明](#随机情况说明)
 - [ModelZoo主页](#ModelZoo主页)
 
@@ -50,8 +51,8 @@
 - 框架
     - [MindSpore](https://www.mindspore.cn/install)
 - 如需查看详情，请参见如下资源：
-    - [MindSpore教程](https://www.mindspore.cn/tutorials/zh-CN/master/index.html)
-    - [MindSpore Python API](https://www.mindspore.cn/docs/api/zh-CN/master/index.html)
+    - [MindSpore教程](https://www.mindspore.cn/tutorial/training/zh-CN/master/index.html)
+    - [MindSpore Python API](https://www.mindspore.cn/doc/api_python/zh-CN/master/index.html)
 
 # 快速入门
 
@@ -75,25 +76,35 @@ sh scripts/run_eval.sh
 ```path
 └── IBNNet  
  ├── README.md                           // IBNNet相关描述
- ├── scripts  
-  ├── run_distribute_train.sh    // 用于分布式训练的shell脚本
+ ├── ascend310_infer                     //310推理
+  ├── inc
+   ├── utils.h
+  ├── src
+   ├── main.cc
+   ├── utils.cc
+  ├── build.sh
+  └── CMakeLists.txt
+ ├── scripts
+  ├── run_310_infer.sh               // 用于310推理的shell脚本
+  ├── run_distribute_train.sh        // 用于分布式训练的shell脚本
   ├── run_distribute_train_gpu.sh    // 用于GPU分布式训练的shell脚本
-  ├── run_standalone_train.sh    // 用于单机训练的shell脚本
-  ├── run_standalone_train.sh    // 用于GPU单机训练的shell脚本
-  ├── run_eval.sh     // 用于评估的shell脚本
-  └── run_eval.sh     // 用于GPU评估的shell脚本
+  ├── run_standalone_train.sh        // 用于单机训练的shell脚本
+  ├── run_standalone_train.sh        // 用于GPU单机训练的shell脚本
+  ├── run_eval.sh                    // 用于评估的shell脚本
+  └── run_eval.sh                    // 用于GPU评估的shell脚本
  ├── src
-  ├── loss.py                       //损失函数
+  ├── loss.py                         //损失函数
   ├── lr_generator.py                 //生成学习率
   ├── config.py                       // 参数配置
   ├── dataset.py                      // 创建数据集
-  ├── resnet_ibn.py                  // IBNNet架构
+  ├── resnet_ibn.py                   // IBNNet架构
  ├── utils
   ├── pth2ckpt.py                       //转换pth文件为ckpt文件
  ├── export.py
  ├── eval.py                             // 测试脚本
  ├── train.py                            // 训练脚本
-
+ ├── preprocess.py                       // 310推理数据预处理
+ ├── preprocess.py                       // 310推理数据后处理
 
 ```
 
@@ -191,11 +202,36 @@ sh scripts/run_eval_gpu.sh path/evalset path/ckpt
 ============== Accuracy:{'top_5_accuracy': 0.93684, 'top_1_accuracy': 0.7743} ==============
 ```
 
+## 导出mindir模型
+
+```python
+python export.py --ckpt_file [CKPT_PATH] --file_name [FILE_NAME] --file_format [FILE_FORMAT]
+```
+
+参数`ckpt_file` 是必需的，`EXPORT_FORMAT` 必须在 ["AIR", "MINDIR"]中进行选择。
+
+# 推理过程
+
+## 用法
+
+在执行推理之前，需要通过export.py导出mindir文件。
+
+```bash
+# Ascend310 推理
+bash run_310_infer.sh [MINDIR_PATH] [DATASET_PATH]
+```
+
+`MINDIR_PATH` 为mindir文件路径，`DATASET_PATH` 表示数据集路径。
+
+### 结果
+
+推理结果保存在当前路径，可在acc.log中看到最终精度结果。
+
 # 模型描述
 
 ## 性能
 
-### 评估性能
+### 训练性能
 
 | 参数          | IBN-Net                                         |
 | ------------- | ----------------------------------------------- |
@@ -215,7 +251,7 @@ sh scripts/run_eval_gpu.sh path/evalset path/ckpt
 | 微调检查点 | 293M （.ckpt file） |
 | 脚本 | [脚本路径](https://gitee.com/mindspore/models/tree/master/research/cv/ibnnet) |
 
-### 推理性能
+### 评估性能
 
 | 参数          | IBN-Net            |
 | ------------- | ------------------ |
@@ -227,55 +263,6 @@ sh scripts/run_eval_gpu.sh path/evalset path/ckpt
 | 输出          | 概率               |
 | 准确性        | 1卡：77.45%; 8卡：77.45% |
 
-## 使用方法
-
-### 推理
-
-如果您需要使用已训练模型在GPU、Ascend 910、Ascend 310等多个硬件平台上进行推理，可参考[此处](https://www.mindspore.cn/docs/programming_guide/zh-CN/master/multi_platform_inference.html)。操作示例如下：
-
-```python
-# 加载未知数据集进行推理
-dataset = dataset.create_dataset(cfg.data_path, 1, False)
-
-# 定义模型
-net = resnet50_ibn_a(num_classes=1000, pretrained=False)
-param_dict = load_checkpoint(args.ckpt_url)
-load_param_into_net(net, param_dict)
-print('Load Pretrained parameters done!')
-
-criterion = SoftmaxCrossEntropyExpand(sparse=True)
-
-step = train_dataset.get_dataset_size()
-lr = lr_generator(args.lr, train_epoch, steps_per_epoch=step)
-optimizer = nn.SGD(params=net.trainable_params(), learning_rate=lr,
-momentum=args.momentum, weight_decay=args.weight_decay)
-
-# 模型变形
-model = Model(net, loss_fn=criterion, optimizer=optimizer, metrics={"Accuracy": Accuracy()})
-
-time_cb = TimeMonitor(data_size=train_dataset.get_dataset_size())
-loss_cb = LossMonitor()
-
-# 设置并应用检查点参数
-config_ck = CheckpointConfig(save_checkpoint_steps=step, keep_checkpoint_max=5)
-ckpoint_cb = ModelCheckpoint(prefix="ResNet50_" + str(device_id), config=config_ck, directory='/cache/train_output/device_' + str(device_id))
-
-cb = [ckpoint_cb, time_cb, loss_cb, eval_cb]
-model.train(train_epoch, train_dataset, callbacks=cb)
-
-# 加载预训练模型
-param_dict = load_checkpoint(cfg.checkpoint_path)
-load_param_into_net(net, param_dict)
-
-# 对未知数据集进行预测
-acc = model.eval(eval_dataset)
-print("accuracy: ", acc)
-```
-
-### 迁移学习
-
-待补充
-
 # 随机情况说明
 
 在dataset.py中，我们设置了“create_dataset_ImageNet”函数内的种子。
@@ -283,3 +270,4 @@ print("accuracy: ", acc)
 # ModelZoo主页  
 
  请浏览官网[主页](https://gitee.com/mindspore/models)。
+
diff --git a/research/cv/ibnnet/ascend310_infer/CMakeLists.txt b/research/cv/ibnnet/ascend310_infer/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..ee3c85447340e0449ff2b70ed24f60a17e07b2b6
--- /dev/null
+++ b/research/cv/ibnnet/ascend310_infer/CMakeLists.txt
@@ -0,0 +1,14 @@
+cmake_minimum_required(VERSION 3.14.1)
+project(Ascend310Infer)
+add_compile_definitions(_GLIBCXX_USE_CXX11_ABI=0)
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O0 -g -std=c++17 -Werror -Wall -fPIE -Wl,--allow-shlib-undefined")
+set(PROJECT_SRC_ROOT ${CMAKE_CURRENT_LIST_DIR}/)
+option(MINDSPORE_PATH "mindspore install path" "")
+include_directories(${MINDSPORE_PATH})
+include_directories(${MINDSPORE_PATH}/include)
+include_directories(${PROJECT_SRC_ROOT})
+find_library(MS_LIB libmindspore.so ${MINDSPORE_PATH}/lib)
+file(GLOB_RECURSE MD_LIB ${MINDSPORE_PATH}/_c_dataengine*)
+
+add_executable(main src/main.cc src/utils.cc)
+target_link_libraries(main ${MS_LIB} ${MD_LIB} gflags)
diff --git a/research/cv/ibnnet/ascend310_infer/build.sh b/research/cv/ibnnet/ascend310_infer/build.sh
new file mode 100644
index 0000000000000000000000000000000000000000..285514e19f2a1878a7bf8f0eed3c99fbc73868c4
--- /dev/null
+++ b/research/cv/ibnnet/ascend310_infer/build.sh
@@ -0,0 +1,29 @@
+#!/bin/bash
+# Copyright 2021 Huawei Technologies Co., Ltd
+#
+# 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.
+# ============================================================================
+if [ -d out ]; then
+    rm -rf out
+fi
+
+mkdir out
+cd out || exit
+
+if [ -f "Makefile" ]; then
+  make clean
+fi
+
+cmake .. \
+    -DMINDSPORE_PATH="`pip3.7 show mindspore-ascend | grep Location | awk '{print $2"/mindspore"}' | xargs realpath`"
+make
diff --git a/research/cv/ibnnet/ascend310_infer/inc/utils.h b/research/cv/ibnnet/ascend310_infer/inc/utils.h
new file mode 100644
index 0000000000000000000000000000000000000000..f8ae1e5b473d869b77af8d725a280d7c7665527c
--- /dev/null
+++ b/research/cv/ibnnet/ascend310_infer/inc/utils.h
@@ -0,0 +1,35 @@
+/**
+ * Copyright 2021 Huawei Technologies Co., Ltd
+ * 
+ * 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.
+ */
+
+#ifndef MINDSPORE_INFERENCE_UTILS_H_
+#define MINDSPORE_INFERENCE_UTILS_H_
+
+#include <sys/stat.h>
+#include <dirent.h>
+#include <vector>
+#include <string>
+#include <memory>
+#include "include/api/types.h"
+
+std::vector<std::string> GetAllFiles(std::string_view dirName);
+DIR *OpenDir(std::string_view dirName);
+std::string RealPath(std::string_view path);
+mindspore::MSTensor ReadFileToTensor(const std::string &file);
+int WriteResult(const std::string& imageFile, const std::vector<mindspore::MSTensor> &outputs);
+std::vector<std::string> GetAllFiles(std::string dir_name);
+std::vector<std::vector<std::string>> GetAllInputData(std::string dir_name);
+
+#endif
diff --git a/research/cv/ibnnet/ascend310_infer/src/main.cc b/research/cv/ibnnet/ascend310_infer/src/main.cc
new file mode 100644
index 0000000000000000000000000000000000000000..1b1f2a91f01ef99b8dc63ef5d201fa3219e5d141
--- /dev/null
+++ b/research/cv/ibnnet/ascend310_infer/src/main.cc
@@ -0,0 +1,152 @@
+/**
+ * Copyright 2021 Huawei Technologies Co., Ltd
+ * 
+ * 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 <sys/time.h>
+#include <gflags/gflags.h>
+#include <dirent.h>
+#include <iostream>
+#include <string>
+#include <algorithm>
+#include <iosfwd>
+#include <vector>
+#include <fstream>
+#include <sstream>
+
+#include "include/api/model.h"
+#include "include/api/context.h"
+#include "include/api/types.h"
+#include "include/api/serialization.h"
+#include "include/dataset/vision_ascend.h"
+#include "include/dataset/execute.h"
+#include "include/dataset/transforms.h"
+#include "include/dataset/vision.h"
+#include "inc/utils.h"
+
+using mindspore::Context;
+using mindspore::Serialization;
+using mindspore::Model;
+using mindspore::Status;
+using mindspore::ModelType;
+using mindspore::GraphCell;
+using mindspore::kSuccess;
+using mindspore::MSTensor;
+using mindspore::dataset::Execute;
+using mindspore::dataset::vision::Decode;
+using mindspore::dataset::vision::Resize;
+using mindspore::dataset::vision::CenterCrop;
+using mindspore::dataset::vision::Normalize;
+using mindspore::dataset::vision::HWC2CHW;
+
+
+DEFINE_string(mindir_path, "", "mindir path");
+DEFINE_string(dataset_name, "imagenet2012", "['cifar10', 'imagenet2012']");
+DEFINE_string(input0_path, ".", "input0 path");
+DEFINE_int32(device_id, 0, "device id");
+
+int load_model(Model *model, std::vector<MSTensor> *model_inputs, std::string mindir_path, int device_id) {
+  if (RealPath(mindir_path).empty()) {
+    std::cout << "Invalid mindir" << std::endl;
+    return 1;
+  }
+
+  auto context = std::make_shared<Context>();
+  auto ascend310 = std::make_shared<mindspore::Ascend310DeviceInfo>();
+  ascend310->SetDeviceID(device_id);
+  context->MutableDeviceInfo().push_back(ascend310);
+  mindspore::Graph graph;
+  Serialization::Load(mindir_path, ModelType::kMindIR, &graph);
+
+  Status ret = model->Build(GraphCell(graph), context);
+  if (ret != kSuccess) {
+    std::cout << "ERROR: Build failed." << std::endl;
+    return 1;
+  }
+
+  *model_inputs = model->GetInputs();
+  if (model_inputs->empty()) {
+    std::cout << "Invalid model, inputs is empty." << std::endl;
+    return 1;
+  }
+  return 0;
+}
+
+int main(int argc, char **argv) {
+  gflags::ParseCommandLineFlags(&argc, &argv, true);
+
+  Model model;
+  std::vector<MSTensor> model_inputs;
+  load_model(&model, &model_inputs, FLAGS_mindir_path, FLAGS_device_id);
+
+  std::map<double, double> costTime_map;
+  struct timeval start = {0};
+  struct timeval end = {0};
+  double startTimeMs;
+  double endTimeMs;
+
+  auto input0_files = GetAllInputData(FLAGS_input0_path);
+  if (input0_files.empty()) {
+    std::cout << "ERROR: no input data." << std::endl;
+    return 1;
+  }
+  size_t size = input0_files.size();
+  for (size_t i = 0; i < size; ++i) {
+    for (size_t j = 0; j < input0_files[i].size(); ++j) {
+      std::vector<MSTensor> inputs;
+      std::vector<MSTensor> outputs;
+      std::cout << "Start predict input files:" << input0_files[i][j] <<std::endl;
+      auto decode = Decode();
+      auto resize = Resize({256, 256});
+      auto centercrop = CenterCrop({224, 224});
+      auto normalize = Normalize({123.675, 116.28, 103.53}, {58.395, 57.12, 57.375});
+      auto hwc2chw = HWC2CHW();
+
+      Execute SingleOp({decode, resize, centercrop, normalize, hwc2chw});
+      auto imgDvpp = std::make_shared<MSTensor>();
+      SingleOp(ReadFileToTensor(input0_files[i][j]), imgDvpp.get());
+      inputs.emplace_back(model_inputs[0].Name(), model_inputs[0].DataType(), model_inputs[0].Shape(),
+                          imgDvpp->Data().get(), imgDvpp->DataSize());
+    gettimeofday(&start, nullptr);
+    Status ret = model.Predict(inputs, &outputs);
+    gettimeofday(&end, nullptr);
+    if (ret != kSuccess) {
+      std::cout << "Predict " << input0_files[i][j] << " failed." << std::endl;
+      return 1;
+    }
+    startTimeMs = (1.0 * start.tv_sec * 1000000 + start.tv_usec) / 1000;
+    endTimeMs = (1.0 * end.tv_sec * 1000000 + end.tv_usec) / 1000;
+    costTime_map.insert(std::pair<double, double>(startTimeMs, endTimeMs));
+    WriteResult(input0_files[i][j], outputs);
+    }
+  }
+  double average = 0.0;
+  int inferCount = 0;
+
+  for (auto iter = costTime_map.begin(); iter != costTime_map.end(); iter++) {
+    double diff = 0.0;
+    diff = iter->second - iter->first;
+    average += diff;
+    inferCount++;
+  }
+  average = average / inferCount;
+  std::stringstream timeCost;
+  timeCost << "NN inference cost average time: "<< average << " ms of infer_count " << inferCount << std::endl;
+  std::cout << "NN inference cost average time: "<< average << "ms of infer_count " << inferCount << std::endl;
+  std::string fileName = "./time_Result" + std::string("/test_perform_static.txt");
+  std::ofstream fileStream(fileName.c_str(), std::ios::trunc);
+  fileStream << timeCost.str();
+  fileStream.close();
+  costTime_map.clear();
+  return 0;
+}
diff --git a/research/cv/ibnnet/ascend310_infer/src/utils.cc b/research/cv/ibnnet/ascend310_infer/src/utils.cc
new file mode 100644
index 0000000000000000000000000000000000000000..d71f388b83d23c2813d8bfc883dbcf2e7e0e4ef0
--- /dev/null
+++ b/research/cv/ibnnet/ascend310_infer/src/utils.cc
@@ -0,0 +1,185 @@
+/**
+ * Copyright 2021 Huawei Technologies Co., Ltd
+ * 
+ * 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 <fstream>
+#include <algorithm>
+#include <iostream>
+#include "inc/utils.h"
+
+using mindspore::MSTensor;
+using mindspore::DataType;
+
+
+std::vector<std::vector<std::string>> GetAllInputData(std::string dir_name) {
+  std::vector<std::vector<std::string>> ret;
+
+  DIR *dir = OpenDir(dir_name);
+  if (dir == nullptr) {
+    return {};
+  }
+  struct dirent *filename;
+  /* read all the files in the dir ~ */
+  std::vector<std::string> sub_dirs;
+  while ((filename = readdir(dir)) != nullptr) {
+    std::string d_name = std::string(filename->d_name);
+    // get rid of "." and ".."
+    if (d_name == "." || d_name == ".." || d_name.empty()) {
+      continue;
+    }
+    std::string dir_path = RealPath(std::string(dir_name) + "/" + filename->d_name);
+    struct stat s;
+    lstat(dir_path.c_str(), &s);
+    if (!S_ISDIR(s.st_mode)) {
+      continue;
+    }
+
+    sub_dirs.emplace_back(dir_path);
+  }
+  std::sort(sub_dirs.begin(), sub_dirs.end());
+
+  (void)std::transform(sub_dirs.begin(), sub_dirs.end(), std::back_inserter(ret),
+                       [](const std::string &d) { return GetAllFiles(d); });
+
+  return ret;
+}
+
+
+std::vector<std::string> GetAllFiles(std::string dir_name) {
+  struct dirent *filename;
+  DIR *dir = OpenDir(dir_name);
+  if (dir == nullptr) {
+    return {};
+  }
+
+  std::vector<std::string> res;
+  while ((filename = readdir(dir)) != nullptr) {
+    std::string d_name = std::string(filename->d_name);
+    if (d_name == "." || d_name == ".." || d_name.size() <= 3) {
+      continue;
+    }
+    res.emplace_back(std::string(dir_name) + "/" + filename->d_name);
+  }
+  std::sort(res.begin(), res.end());
+
+  return res;
+}
+
+
+std::vector<std::string> GetAllFiles(std::string_view dirName) {
+  struct dirent *filename;
+  DIR *dir = OpenDir(dirName);
+  if (dir == nullptr) {
+    return {};
+  }
+  std::vector<std::string> res;
+  while ((filename = readdir(dir)) != nullptr) {
+    std::string dName = std::string(filename->d_name);
+    if (dName == "." || dName == ".." || filename->d_type != DT_REG) {
+      continue;
+    }
+    res.emplace_back(std::string(dirName) + "/" + filename->d_name);
+  }
+  std::sort(res.begin(), res.end());
+  for (auto &f : res) {
+    std::cout << "image file: " << f << std::endl;
+  }
+  return res;
+}
+
+
+int WriteResult(const std::string& imageFile, const std::vector<MSTensor> &outputs) {
+  std::string homePath = "./result_Files";
+  for (size_t i = 0; i < outputs.size(); ++i) {
+    size_t outputSize;
+    std::shared_ptr<const void> netOutput;
+    netOutput = outputs[i].Data();
+    outputSize = outputs[i].DataSize();
+    int pos = imageFile.rfind('/');
+    std::string fileName(imageFile, pos + 1);
+    fileName.replace(fileName.find('.'), fileName.size() - fileName.find('.'), '_' + std::to_string(i) + ".bin");
+    std::string outFileName = homePath + "/" + fileName;
+    FILE *outputFile = fopen(outFileName.c_str(), "wb");
+    fwrite(netOutput.get(), outputSize, sizeof(char), outputFile);
+    fclose(outputFile);
+    outputFile = nullptr;
+  }
+  return 0;
+}
+
+mindspore::MSTensor ReadFileToTensor(const std::string &file) {
+  if (file.empty()) {
+    std::cout << "Pointer file is nullptr" << std::endl;
+    return mindspore::MSTensor();
+  }
+
+  std::ifstream ifs(file);
+  if (!ifs.good()) {
+    std::cout << "File: " << file << " is not exist" << std::endl;
+    return mindspore::MSTensor();
+  }
+
+  if (!ifs.is_open()) {
+    std::cout << "File: " << file << "open failed" << std::endl;
+    return mindspore::MSTensor();
+  }
+
+  ifs.seekg(0, std::ios::end);
+  size_t size = ifs.tellg();
+  mindspore::MSTensor buffer(file, mindspore::DataType::kNumberTypeUInt8, {static_cast<int64_t>(size)}, nullptr, size);
+
+  ifs.seekg(0, std::ios::beg);
+  ifs.read(reinterpret_cast<char *>(buffer.MutableData()), size);
+  ifs.close();
+
+  return buffer;
+}
+
+
+DIR *OpenDir(std::string_view dirName) {
+  if (dirName.empty()) {
+    std::cout << " dirName is null ! " << std::endl;
+    return nullptr;
+  }
+  std::string realPath = RealPath(dirName);
+  struct stat s;
+  lstat(realPath.c_str(), &s);
+  if (!S_ISDIR(s.st_mode)) {
+    std::cout << "dirName is not a valid directory !" << std::endl;
+    return nullptr;
+  }
+  DIR *dir;
+  dir = opendir(realPath.c_str());
+  if (dir == nullptr) {
+    std::cout << "Can not open dir " << dirName << std::endl;
+    return nullptr;
+  }
+  std::cout << "Successfully opened the dir " << dirName << std::endl;
+  return dir;
+}
+
+std::string RealPath(std::string_view path) {
+  char realPathMem[PATH_MAX] = {0};
+  char *realPathRet = nullptr;
+  realPathRet = realpath(path.data(), realPathMem);
+  if (realPathRet == nullptr) {
+    std::cout << "File: " << path << " is not exist.";
+    return "";
+  }
+
+  std::string realPath(realPathMem);
+  std::cout << path << " realpath is: " << realPath << std::endl;
+  return realPath;
+}
diff --git a/research/cv/ibnnet/eval.py b/research/cv/ibnnet/eval.py
index 69733a27852552a7d62d46ecec3c65cf21fc47f4..b619806b7d056a6024c386926c037084deecba43 100644
--- a/research/cv/ibnnet/eval.py
+++ b/research/cv/ibnnet/eval.py
@@ -59,7 +59,7 @@ if __name__ == "__main__":
     step = 60
     target = args.device_target
     context.set_context(mode=context.GRAPH_MODE, device_target=target, save_graphs=False)
-    context.set_context(device_id=args.device_id)
+    context.set_context(device_id=args.device_id, enable_auto_mixed_precision=True)
 
     lr = lr_generator(cfg.lr, train_epoch, steps_per_epoch=step)
     net = resnet50_ibn_a(num_classes=cfg.class_num)
diff --git a/research/cv/ibnnet/postprocess.py b/research/cv/ibnnet/postprocess.py
new file mode 100644
index 0000000000000000000000000000000000000000..4144a3ee71e41825744a5651d01e04983513e355
--- /dev/null
+++ b/research/cv/ibnnet/postprocess.py
@@ -0,0 +1,48 @@
+# Copyright 2021 Huawei Technologies Co., Ltd
+#
+# 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.
+# ============================================================================
+"""postprocess for 310 inference"""
+import os
+import argparse
+import json
+import numpy as np
+from mindspore.nn import Top1CategoricalAccuracy, Top5CategoricalAccuracy
+parser = argparse.ArgumentParser(description="postprocess")
+label_path = "./preprocess_Result/cifar10_label_ids.npy"
+parser.add_argument("--result_dir", type=str, default="./result_Files", help="result files path.")
+parser.add_argument('--dataset_name', type=str, default="imagenet2012")
+parser.add_argument("--label_dir", type=str, default=label_path, help="image file path.")
+args = parser.parse_args()
+
+def calcul_acc(lab, preds):
+    return sum(1 for x, y in zip(lab, preds) if x == y) / len(lab)
+if __name__ == '__main__':
+    batch_size = 1
+    top1_acc = Top1CategoricalAccuracy()
+    rst_path = args.result_dir
+    label_list = []
+    pred_list = []
+    file_list = os.listdir(rst_path)
+    top5_acc = Top5CategoricalAccuracy()
+    with open('./preprocess_Result/imagenet_label.json', "r") as label:
+        labels = json.load(label)
+    for f in file_list:
+        label = f.split("_0.bin")[0] + ".JPEG"
+        label_list.append(labels[label])
+        pred = np.fromfile(os.path.join(rst_path, f), np.float32)
+        pred = pred.reshape(batch_size, int(pred.shape[0] / batch_size))
+        top1_acc.update(pred, [labels[label],])
+        top5_acc.update(pred, [labels[label],])
+    print("Top1 acc: ", top1_acc.eval())
+    print("Top5 acc: ", top5_acc.eval())
diff --git a/research/cv/ibnnet/preprocess.py b/research/cv/ibnnet/preprocess.py
new file mode 100644
index 0000000000000000000000000000000000000000..da3a7f8f5ebf3b7645f493246bb5181bbe5990dd
--- /dev/null
+++ b/research/cv/ibnnet/preprocess.py
@@ -0,0 +1,48 @@
+# Copyright 2021 Huawei Technologies Co., Ltd
+#
+# 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.
+# ============================================================================
+"""preprocess"""
+import os
+import argparse
+import json
+parser = argparse.ArgumentParser('preprocess')
+parser.add_argument('--dataset_name', type=str, choices=["cifar10", "imagenet2012"], default="imagenet2012")
+parser.add_argument('--data_path', type=str, default='', help='eval data dir')
+parser.add_argument('--result_path', type=str, default='./preprocess_Result/', help='result path')
+def create_label(result_path, dir_path):
+    """create label json for imagenet"""
+    print("[WARNING] Create imagenet label. Currently only use for Imagenet2012!")
+    dirs = os.listdir(dir_path)
+    file_list = []
+    for file in dirs:
+        file_list.append(file)
+    file_list = sorted(file_list)
+
+    total = 0
+    img_label = {}
+    for i, file_dir in enumerate(file_list):
+        files = os.listdir(os.path.join(dir_path, file_dir))
+        for f in files:
+            img_label[f] = i
+        total += len(files)
+
+    json_file = os.path.join(result_path, "imagenet_label.json")
+    with open(json_file, "w+") as label:
+        json.dump(img_label, label)
+
+    print("[INFO] Completed! Total {} data.".format(total))
+
+args = parser.parse_args()
+if __name__ == "__main__":
+    create_label('./preprocess_Result/', args.data_path)
diff --git a/research/cv/ibnnet/scripts/run_310_infer.sh b/research/cv/ibnnet/scripts/run_310_infer.sh
new file mode 100644
index 0000000000000000000000000000000000000000..acf73c1fcd99ef81036e3fbbcce00e75536e1412
--- /dev/null
+++ b/research/cv/ibnnet/scripts/run_310_infer.sh
@@ -0,0 +1,124 @@
+#!/bin/bash
+# Copyright 2021 Huawei Technologies Co., Ltd
+#
+# 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.
+# ============================================================================
+
+if [[ $# -lt 2 || $# -gt 3 ]]; then
+    echo "Usage: bash run_infer_310.sh [MINDIR_PATH] [DATASET_PATH] [DEVICE_ID]
+    DEVICE_ID is optional, it can be set by environment variable device_id, otherwise the value is zero"
+exit 1
+fi
+
+get_real_path(){
+    if [ "${1:0:1}" == "/" ]; then
+        echo "$1"
+    else
+        echo "$(realpath -m $PWD/$1)"
+    fi
+}
+model=$(get_real_path $1)
+dataset_name='imagenet2012'
+dataset_path=$(get_real_path $2)
+need_preprocess='y'
+device_id=0
+if [ $# == 3 ]; then
+    device_id=$3
+fi
+
+echo "mindir name: "$model
+echo "dataset name: "$dataset_name
+echo "dataset path: "$dataset_path
+echo "need preprocess: "$need_preprocess
+echo "device id: "$device_id
+
+export ASCEND_HOME=/usr/local/Ascend/
+if [ -d ${ASCEND_HOME}/ascend-toolkit ]; then
+    export PATH=$ASCEND_HOME/fwkacllib/bin:$ASCEND_HOME/fwkacllib/ccec_compiler/bin:$ASCEND_HOME/ascend-toolkit/latest/fwkacllib/ccec_compiler/bin:$ASCEND_HOME/ascend-toolkit/latest/atc/bin:$PATH
+    export LD_LIBRARY_PATH=$ASCEND_HOME/fwkacllib/lib64:/usr/local/lib:$ASCEND_HOME/ascend-toolkit/latest/atc/lib64:$ASCEND_HOME/ascend-toolkit/latest/fwkacllib/lib64:$ASCEND_HOME/driver/lib64:$ASCEND_HOME/add-ons:$LD_LIBRARY_PATH
+    export TBE_IMPL_PATH=$ASCEND_HOME/ascend-toolkit/latest/opp/op_impl/built-in/ai_core/tbe
+    export PYTHONPATH=$ASCEND_HOME/fwkacllib/python/site-packages:${TBE_IMPL_PATH}:$ASCEND_HOME/ascend-toolkit/latest/fwkacllib/python/site-packages:$PYTHONPATH
+    export ASCEND_OPP_PATH=$ASCEND_HOME/ascend-toolkit/latest/opp
+else
+    export PATH=$ASCEND_HOME/fwkacllib/bin:$ASCEND_HOME/fwkacllib/ccec_compiler/bin:$ASCEND_HOME/atc/ccec_compiler/bin:$ASCEND_HOME/atc/bin:$PATH
+    export LD_LIBRARY_PATH=$ASCEND_HOME/fwkacllib/lib64:/usr/local/lib:$ASCEND_HOME/atc/lib64:$ASCEND_HOME/acllib/lib64:$ASCEND_HOME/driver/lib64:$ASCEND_HOME/add-ons:$LD_LIBRARY_PATH
+    export PYTHONPATH=$ASCEND_HOME/fwkacllib/python/site-packages:$ASCEND_HOME/atc/python/site-packages:$PYTHONPATH
+    export ASCEND_OPP_PATH=$ASCEND_HOME/opp
+fi
+export ASCEND_HOME=/usr/local/Ascend
+export PATH=$ASCEND_HOME/fwkacllib/ccec_compiler/bin:$ASCEND_HOME/fwkacllib/bin:$ASCEND_HOME/toolkit/bin:$PATH
+export LD_LIBRARY_PATH=/usr/local/lib/:/usr/local/fwkacllib/lib64:$ASCEND_HOME/driver/lib64:$ASCEND_HOME/add-ons:/usr/local/Ascend/toolkit/lib64:$LD_LIBRARY_PATH
+export PYTHONPATH=$ASCEND_HOME/fwkacllib/python/site-packages
+export PATH=/usr/local/python375/bin:$PATH
+export NPU_HOST_LIB=/usr/local/Ascend/acllib/lib64/stub
+export ASCEND_OPP_PATH=/usr/local/Ascend/opp
+export ASCEND_AICPU_PATH=/usr/local/Ascend
+export LD_LIBRARY_PATH=/usr/local/lib64/:$LD_LIBRARY_PATH
+function preprocess_data()
+{
+    if [ -d preprocess_Result ]; then
+        rm -rf ./preprocess_Result
+    fi
+    mkdir preprocess_Result
+    python3.7 ../preprocess.py --dataset_name=$dataset_name --data_path=$dataset_path 
+}
+
+function compile_app()
+{
+    cd ../ascend310_infer/ || exit
+    bash build.sh &> build.log
+}
+
+function infer()
+{
+    cd - || exit
+    if [ -d result_Files ]; then
+        rm -rf ./result_Files
+    fi
+    if [ -d time_Result ]; then
+        rm -rf ./time_Result
+    fi
+    mkdir result_Files
+    mkdir time_Result
+
+    ../ascend310_infer/out/main --mindir_path=$model --dataset_name=$dataset_name --input0_path=$dataset_path --device_id=$device_id  &> infer.log
+
+}
+
+function cal_acc()
+{
+    python3.7 ../postprocess.py --dataset_name=$dataset_name  &> acc.log
+}
+
+if [ $need_preprocess == "y" ]; then
+    preprocess_data
+    if [ $? -ne 0 ]; then
+        echo "preprocess dataset failed"
+        exit 1
+    fi
+fi
+compile_app
+if [ $? -ne 0 ]; then
+    echo "compile app code failed"
+    exit 1
+fi
+infer
+if [ $? -ne 0 ]; then
+    echo " execute inference failed"
+    exit 1
+fi
+cal_acc
+if [ $? -ne 0 ]; then
+    echo "calculate accuracy failed"
+    exit 1
+fi
diff --git a/research/cv/ibnnet/train.py b/research/cv/ibnnet/train.py
index 15abb4796eed9a248dc000a8efd437a33527fc6b..a3eb7ac9fba9b89b8268921b798c1d40c147bdee 100644
--- a/research/cv/ibnnet/train.py
+++ b/research/cv/ibnnet/train.py
@@ -20,8 +20,7 @@ import os
 
 import mindspore.nn as nn
 from mindspore import context
-from mindspore.context import ParallelMode
-from mindspore.train.model import Model
+from mindspore.train.model import Model, ParallelMode
 from mindspore.train.serialization import load_checkpoint, load_param_into_net
 from mindspore.train.callback import ModelCheckpoint, CheckpointConfig, LossMonitor, TimeMonitor, Callback
 from mindspore.nn.metrics import Accuracy
@@ -97,7 +96,8 @@ if __name__ == "__main__":
     if args.device_num > 1:
         if target == 'Ascend':
             device_id = int(os.getenv('DEVICE_ID'))
-            context.set_context(device_id=device_id)
+            context.set_context(device_id=device_id,
+                                enable_auto_mixed_precision=True)
             context.set_auto_parallel_context(parallel_mode=ParallelMode.DATA_PARALLEL,
                                               gradients_mean=True,
                                               auto_parallel_search_mode="recursive_programming")