diff --git a/research/cv/AVA_hpa/README.md b/research/cv/AVA_hpa/README.md
index 80a76f4eb528d679cece52cc145c4c6d34ed2a47..6332d965b42156fd962bf68b36b5d0acb16ec91f 100644
--- a/research/cv/AVA_hpa/README.md
+++ b/research/cv/AVA_hpa/README.md
@@ -11,6 +11,10 @@
     - [Pre-training Process](#pre-training-process)
     - [Training Process](#training-process)
     - [Evaluation](#evaluation)
+    - [Inference Process](#inference-process)
+        - [Export MindIR](#export-mindir)
+        - [Infer on Ascend310](#infer-on-ascend310)
+        - [result](#result)
 
 # [AVA_hpa Description](#contents)
 
@@ -254,3 +258,32 @@ The value of performance will be achieved as follows:
 ```shell
 {'results_return': ( 0.6975821515082009, 0.7734114826162249, 0.9415286419176973)} #macroF1,microF1,auc
 ```
+
+## Inference Process
+
+### [Export MindIR](#contents)
+
+```shell
+python export.py --ckpt_path [CKPT_PATH] --model_arch [MODEL_ARCH] --file_name [FILE_NAME] --file_format [FILE_FORMAT]
+```
+
+- The `CKPT_PATH` parameter is required.
+- `MODEL_ARCH` is model architecture, should be in ['resnet18', 'resnet50', 'resnet101'].
+- `file_format` should be in ["AIR", "MINDIR"]
+
+### Infer on Ascend310
+
+Before performing inference, the mindir file must be exported by `export.py` script. We only provide an example of inference using MINDIR model.
+
+```shell
+# Ascend310 inference
+bash run_infer_310.sh [MINDIR_PATH] [DATASET_PATH] [NEED_PREPROCESS] [DEVICE_ID]
+```
+
+- `DATASET_PATH` is the hpa dataset path.
+- `NEED_PREPROCESS` means weather need preprocess or not, it's value is 'y' or 'n'.
+- `DEVICE_ID` is optional, default value is 0.
+
+### result
+
+Inference result is saved in current path, you can find result like this in acc.log file.
diff --git a/research/cv/AVA_hpa/ascend310_infer/CMakeLists.txt b/research/cv/AVA_hpa/ascend310_infer/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..ee3c85447340e0449ff2b70ed24f60a17e07b2b6
--- /dev/null
+++ b/research/cv/AVA_hpa/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/AVA_hpa/ascend310_infer/build.sh b/research/cv/AVA_hpa/ascend310_infer/build.sh
new file mode 100644
index 0000000000000000000000000000000000000000..c7aac7196661fcfa50c3de063336355bf3831286
--- /dev/null
+++ b/research/cv/AVA_hpa/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="`pip show mindspore-ascend | grep Location | awk '{print $2"/mindspore"}' | xargs realpath`"
+make
\ No newline at end of file
diff --git a/research/cv/AVA_hpa/ascend310_infer/inc/utils.h b/research/cv/AVA_hpa/ascend310_infer/inc/utils.h
new file mode 100644
index 0000000000000000000000000000000000000000..0b400632f51ee34707a5becc00f7f5ba05899b3a
--- /dev/null
+++ b/research/cv/AVA_hpa/ascend310_infer/inc/utils.h
@@ -0,0 +1,33 @@
+/**
+ * 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"
+
+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);
+
+#endif
diff --git a/research/cv/AVA_hpa/ascend310_infer/src/main.cc b/research/cv/AVA_hpa/ascend310_infer/src/main.cc
new file mode 100644
index 0000000000000000000000000000000000000000..39789a5e6c8ed72bd7c13a063ed8bf69fed7ef09
--- /dev/null
+++ b/research/cv/AVA_hpa/ascend310_infer/src/main.cc
@@ -0,0 +1,142 @@
+/**
+ * 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(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);
+
+  auto input0_files = GetAllFiles(FLAGS_input0_path);
+  if (input0_files.empty()) {
+    std::cout << "ERROR: no input data." << std::endl;
+    return 1;
+  }
+  std::map<double, double> costTime_map;
+  struct timeval start = {0};
+  struct timeval end = {0};
+  size_t size = input0_files.size();
+  for (size_t i = 0; i < size; ++i) {
+    double startTimeMs;
+    double endTimeMs;
+    std::vector<MSTensor> inputs;
+    std::vector<MSTensor> outputs;
+    std::cout << "Start predict input files:" << input0_files[i] <<std::endl;
+    auto input0 = ReadFileToTensor(input0_files[i]);
+    inputs.emplace_back(model_inputs[0].Name(), model_inputs[0].DataType(), model_inputs[0].Shape(),
+                        input0.Data().get(), input0.DataSize());
+
+    gettimeofday(&start, nullptr);
+    Status ret = model.Predict(inputs, &outputs);
+    gettimeofday(&end, nullptr);
+    if (ret != kSuccess) {
+      std::cout << "Predict " << input0_files[i] << " 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], 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/AVA_hpa/ascend310_infer/src/utils.cc b/research/cv/AVA_hpa/ascend310_infer/src/utils.cc
new file mode 100644
index 0000000000000000000000000000000000000000..728d57d93624e69f25fc0cd9a3c45afe80dab858
--- /dev/null
+++ b/research/cv/AVA_hpa/ascend310_infer/src/utils.cc
@@ -0,0 +1,145 @@
+/**
+ * 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::string> GetAllFiles(std::string dirName) {
+    struct dirent *filename;
+    DIR *dir = OpenDir(dirName);
+    if (dir == nullptr) {
+        return {};
+    }
+    std::vector<std::string> dirs;
+    std::vector<std::string> files;
+    while ((filename = readdir(dir)) != nullptr) {
+        std::string dName = std::string(filename->d_name);
+        if (dName == "." || dName == "..") {
+            continue;
+        } else if (filename->d_type == DT_DIR) {
+            dirs.emplace_back(std::string(dirName) + "/" + filename->d_name);
+        } else if (filename->d_type == DT_REG) {
+            files.emplace_back(std::string(dirName) + "/" + filename->d_name);
+        } else {
+            continue;
+        }
+    }
+
+    for (auto d : dirs) {
+        dir = OpenDir(d);
+        while ((filename = readdir(dir)) != nullptr) {
+            std::string dName = std::string(filename->d_name);
+            if (dName == "." || dName == ".." || filename->d_type != DT_REG) {
+                continue;
+            }
+            files.emplace_back(std::string(d) + "/" + filename->d_name);
+        }
+    }
+    std::sort(files.begin(), files.end());
+    for (auto &f : files) {
+        std::cout << "image file: " << f << std::endl;
+    }
+    return files;
+}
+
+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/AVA_hpa/export.py b/research/cv/AVA_hpa/export.py
new file mode 100644
index 0000000000000000000000000000000000000000..fa004ae1897c15b09582bf92bb23ba8c65a0e28d
--- /dev/null
+++ b/research/cv/AVA_hpa/export.py
@@ -0,0 +1,67 @@
+# 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.
+# ============================================================================
+"""export"""
+import argparse
+import numpy as np
+
+from mindspore import context, Tensor
+from mindspore.train.serialization import load_checkpoint, load_param_into_net, export
+
+from src.resnet import resnet18, resnet50, resnet101
+from src.network_define_eval import EvalCell310
+
+parser = argparse.ArgumentParser(description="export")
+parser.add_argument("--device_id", type=int, default=0,
+                    help="Device id, default is 0.")
+parser.add_argument("--device_num", type=int, default=1,
+                    help="Use device nums, default is 1.")
+parser.add_argument('--device_target', type=str,
+                    default="Ascend", help='Device target')
+parser.add_argument('--ckpt_path', type=str, default="",
+                    help='model checkpoint path')
+parser.add_argument("--model_arch", type=str, default="resnet18",
+                    choices=['resnet18', 'resnet50', 'resnet101'], help='model architecture')
+parser.add_argument("--classes", type=int, default=10, help='class number')
+parser.add_argument("--file_name", type=str, default="ava_hpa", help='model name')
+parser.add_argument("--file_format", type=str, default="MINDIR",
+                    choices=['AIR', 'MINDIR'], help='model format')
+
+args_opt = parser.parse_args()
+
+if __name__ == "__main__":
+    context.set_context(mode=context.GRAPH_MODE,
+                        device_target=args_opt.device_target)
+    if args_opt.device_target == "Ascend":
+        context.set_context(device_id=args_opt.device_id)
+    ckpt_path = args_opt.ckpt_path
+
+    if args_opt.model_arch == 'resnet18':
+        resnet = resnet18(pretrain=False, classes=args_opt.classes)
+    elif args_opt.model_arch == 'resnet50':
+        resnet = resnet50(pretrain=False, classes=args_opt.classes)
+    elif args_opt.model_arch == 'resnet101':
+        resnet = resnet101(pretrain=False, classes=args_opt.classes)
+    else:
+        raise "Unsupported net work!"
+    param_dict = load_checkpoint(args_opt.ckpt_path)
+    load_param_into_net(resnet, param_dict)
+
+    bag_size_for_eval = 20
+    image_shape = (224, 224)
+    input_shape = (bag_size_for_eval, 3) + image_shape
+
+    test_network = EvalCell310(resnet)
+    input_data0 = Tensor(np.random.uniform(low=0, high=1.0, size=input_shape).astype(np.float32))
+    export(test_network, input_data0, file_name=args_opt.file_name, file_format=args_opt.file_format)
diff --git a/research/cv/AVA_hpa/postprocess.py b/research/cv/AVA_hpa/postprocess.py
new file mode 100644
index 0000000000000000000000000000000000000000..86c819bb757f406ef994477712009f5e95636a78
--- /dev/null
+++ b/research/cv/AVA_hpa/postprocess.py
@@ -0,0 +1,51 @@
+# 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."""
+import os
+import argparse
+import numpy as np
+
+from mindspore import Tensor
+from src.network_define_eval import EvalMetric
+
+parser = argparse.ArgumentParser(description="postprocess")
+parser.add_argument("--result_dir", type=str, default="./result_Files",
+                    help="infer result dataset directory")
+parser.add_argument("--label_dir", type=str, default="",
+                    help="label data file")
+parser.add_argument("--nslice_dir", type=str, default="",
+                    help="nslice data file")
+parser.add_argument("--save_eval_path", type=str, default="./eval_result",
+                    help="eval result path")
+parser.add_argument("--classes", type=int, default=10, help='class number')
+args_opt = parser.parse_args()
+
+if __name__ == '__main__':
+    batch_size = 1
+    bag_size_for_eval = 20
+    acc = EvalMetric(path=args_opt.save_eval_path)
+    result_num = len(os.listdir(args_opt.result_dir))
+    label_list = np.load(args_opt.label_dir)
+    nslice_list = np.load(args_opt.nslice_dir)
+    for i in range(result_num):
+        f = "ava_bs" + str(batch_size) + "_" + str(i) + "_0.bin"
+        feature = np.fromfile(os.path.join(args_opt.result_dir, f), np.float32)
+        feature = Tensor(feature.reshape(bag_size_for_eval, args_opt.classes))
+        label = Tensor(label_list[i])
+        nslice = Tensor(nslice_list[i])
+        inputs = (feature, label, nslice)
+        acc.update(*inputs)
+    result_return = acc.eval()
+    print("The result is {}.".format(result_return))
diff --git a/research/cv/AVA_hpa/preprocess.py b/research/cv/AVA_hpa/preprocess.py
new file mode 100644
index 0000000000000000000000000000000000000000..0f926220f580d6b28f9a57878b122842de358b52
--- /dev/null
+++ b/research/cv/AVA_hpa/preprocess.py
@@ -0,0 +1,53 @@
+# 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 numpy as np
+
+from src.datasets import makeup_dataset
+
+
+parser = argparse.ArgumentParser(description="preprocess")
+parser.add_argument("--data_dir", type=str, default="", help="dataset path")
+parser.add_argument("--classes", type=int, default=10, help='class number')
+parser.add_argument("--pre_result_dir", type=str, default="./preprocess_Result",
+                    help="preprocess data path")
+args_opt = parser.parse_args()
+
+if __name__ == '__main__':
+    batch_size = 1
+    data_dir = args_opt.data_dir
+
+    test_dataset = makeup_dataset(data_dir=data_dir, mode='test', batch_size=1, bag_size=20, classes=args_opt.classes,
+                                  num_parallel_workers=8)
+    test_dataset.__loop_size__ = 1
+
+    image_path = os.path.join(args_opt.pre_result_dir, "00_data")
+    label_path = os.path.join(args_opt.pre_result_dir, "label.npy")
+    nslice_path = os.path.join(args_opt.pre_result_dir, "nslice.npy")
+    os.makedirs(image_path, exist_ok=True)
+    label_list = []
+    nslice_list = []
+    for i, data in enumerate(test_dataset.create_dict_iterator(output_numpy=True)):
+        file_name = "ava_bs" + str(batch_size) + "_" + str(i) + ".bin"
+        file_path = os.path.join(image_path, file_name)
+        data["imgs"].tofile(file_path)
+        label_list.append(data["labels"])
+        nslice_list.append(data["nslice"])
+    np.save(os.path.join(label_path), label_list)
+    np.save(os.path.join(nslice_path), nslice_list)
+    print("=" * 20, "export bin files finished", "=" * 20)
diff --git a/research/cv/AVA_hpa/scripts/run_infer_310.sh b/research/cv/AVA_hpa/scripts/run_infer_310.sh
new file mode 100644
index 0000000000000000000000000000000000000000..7a044eaf7e40a6e276e1b9cf6a4ad3da16fe5ff8
--- /dev/null
+++ b/research/cv/AVA_hpa/scripts/run_infer_310.sh
@@ -0,0 +1,126 @@
+#!/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 3 || $# -gt 4 ]]; then
+    echo "Usage: bash run_infer_310.sh [MINDIR_PATH] [DATASET_PATH] [NEED_PREPROCESS] [DEVICE_ID]
+    NEED_PREPROCESS means weather need preprocess or not, it's value is 'y' or 'n'.
+    DEVICE_ID is optional, it can be set by environment variable device_id, otherwise the value is zero"
+exit 1
+fi
+
+get_real_path(){
+    if [ "${1:0:1}" == "/" ]; then
+        echo "$1"
+    else
+        echo "$(realpath -m $PWD/$1)"
+    fi
+}
+model=$(get_real_path $1)
+dataset_path=$(get_real_path $2)
+
+if [ $3 == 'y' ] || [ $3 == 'n' ]; then
+  need_preprocess=$3
+else
+  echo "weather need preprocess or not, it's value must be in [y, n]"
+  exit 1
+fi
+
+device_id=0
+if [ $# == 4 ]; then
+    device_id=$4
+fi
+
+echo "mindir name: "$model
+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
+
+function compile_app()
+{
+    cd ../ascend310_infer/ || exit
+    bash build.sh &> build.log
+}
+
+function preprocess_data()
+{
+    if [ -d preprocess_Result ]; then
+        rm -rf ./preprocess_Result
+    fi
+    mkdir preprocess_Result
+    rm -rf enhanced.csv && cp ../enhanced.csv .
+    if [ $? -ne 0 ]; then
+      echo "enhanced.csv missed, please check."
+      exit 1
+    fi
+    python ../preprocess.py --data_dir $dataset_path
+}
+
+function infer()
+{
+    cd - || exit
+    if [ -d result_Files ]; then
+        rm -rf ./result_Files
+    fi
+    if [ -d time_Result ]; then
+        rm -rf ./time_Result
+    fi
+    mkdir result_Files
+    mkdir time_Result
+    ../ascend310_infer/out/main --mindir_path=$model --input0_path=./preprocess_Result/00_data --device_id=$device_id  &> infer.log
+}
+
+function cal_acc()
+{
+   python ../postprocess.py --result_dir=./result_Files --label_dir=./preprocess_Result/label.npy --nslice_dir=./preprocess_Result/nslice.npy &> acc.log
+}
+
+if [ $need_preprocess == "y" ]; then
+  preprocess_data
+  if [ $? -ne 0 ]; then
+    echo "preprocess data 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
\ No newline at end of file
diff --git a/research/cv/AVA_hpa/src/network_define_eval.py b/research/cv/AVA_hpa/src/network_define_eval.py
index 49e45bf0f3d73d5ba0c7852f92163668d2d7c450..8a47153a263300b49318b014b45a7e409981feba 100644
--- a/research/cv/AVA_hpa/src/network_define_eval.py
+++ b/research/cv/AVA_hpa/src/network_define_eval.py
@@ -62,6 +62,15 @@ class EvalCell(nn.Cell):
         outputs = self._network(data)
         return outputs, label, nslice
 
+class EvalCell310(nn.Cell):
+    """eval cell"""
+    def __init__(self, network):
+        super(EvalCell310, self).__init__(auto_prefix=False)
+        self._network = network
+
+    def construct(self, data):
+        outputs = self._network(data)
+        return outputs
 
 class EvalMetric(nn.Metric):
     """eval metric"""