!16895 cnnctc and warpctc add 310 infer - amend lenet and mobilenetv2 readme

From: @chenweitao_295 Reviewed-by: @c_34,@oacjiewen Signed-off-by: @c_34
2021-05-27 10:17:50 +08:00 · 2021-05-27 10:17:50 +08:00 · 1691ab80de
parent b75c694032 c2c8db3a0b
commit 1691ab80de
24 changed files with 1367 additions and 10 deletions
--- a/model_zoo/official/cv/cnnctc/README.md
+++ b/model_zoo/official/cv/cnnctc/README.md
@ -15,10 +15,15 @@
        - [Distributed Training](#distributed-training)
    - [Evaluation Process](#evaluation-process)
        - [Evaluation](#evaluation)
+    - [Inference Process](#inference-process)
+        - [Export MindIR](#export-mindir)
+        - [Infer on Ascend310](#infer-on-ascend310)
+        - [result](#result)
 - [Model Description](#model-description)
    - [Performance](#performance)
+        - [Training Performance](#training-performance)
        - [Evaluation Performance](#evaluation-performance)
-        - [Inference Performance](#evaluation-performance)
+        - [Inference Performance](#inference-performance)
    - [How to use](#how-to-use)
        - [Inference](#inference)
        - [Continue Training on the Pretrained Model](#continue-training-on-the-pretrained-model)
@ -145,7 +150,11 @@ The entire code structure is as following:
    |---README.md    // descriptions about cnnctc
    |---train.py    // train scripts
    |---eval.py    // eval scripts
+    |---export.py    // export scripts
+    |---pstprocess.py    // postprocess scripts
+    |---ascend310_infer    // application for 310 inference
    |---scripts
+        |---run_infer_310.sh    // shell script for infer on ascend310
        |---run_standalone_train_ascend.sh    // shell script for standalone on ascend
        |---run_distribute_train_ascend.sh    // shell script for distributed on ascend
        |---run_eval_ascend.sh    // shell script for eval on ascend
@ -251,6 +260,39 @@ bash scripts/run_eval_ascend.sh $TRAINED_CKPT

 The model will be evaluated on the IIIT dataset, sample results and overall accuracy will be printed.

+## [Inference process](#contents)
+
+### Export MindIR
+
+```shell
+python export.py --ckpt_file [CKPT_PATH] --file_name [FILE_NAME] --file_format [EXPORT_FORMAT]
+```
+
+The ckpt_file parameter is required,
+The file_name parameter is file name after export.
+`EXPORT_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.
+Current batch_size can only be set to 1, modify the parameter `TEST_BATCH_SIZE` in `config.py` to 1 before export the model
+
+```shell
+# Ascend310 inference
+bash run_infer_310.sh [MINDIR_PATH] [DATA_PATH] [DVPP] [DEVICE_ID]
+```
+
+- `DVPP` is mandatory, and must choose from ["DVPP", "CPU"], it's case-insensitive. CNNCTC only support CPU mode .
+- `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.
+
+```bash
+'Accuracy': 0.8546
+```
+
 # [Model Description](#contents)

 ## [Performance](#contents)
@ -286,6 +328,20 @@ The model will be evaluated on the IIIT dataset, sample results and overall accu
 | Accuracy            |  85%  |
 | Model for inference | 675M (.ckpt file)         |

+### Inference Performance
+
+| Parameters          | Ascend                      |
+| ------------------- | --------------------------- |
+| Model Version       | CNNCTC                      |
+| Resource            | Ascend 310; CentOS 3.10     |
+| Uploaded Date       | 19/05/2021 (month/day/year) |
+| MindSpore Version   | 1.2.0                       |
+| Dataset             | IIIT5K                      |
+| batch_size          | 1                           |
+| outputs             | Accuracy                    |
+| Accuracy            | Accuracy=0.8546             |
+| Model for inference | 675M(.ckpt file)            |
+
 ## [How to use](#contents)

 ### Inference
--- a/model_zoo/official/cv/cnnctc/README_CN.md
+++ b/model_zoo/official/cv/cnnctc/README_CN.md
@ -18,10 +18,15 @@
        - [训练结果](#训练结果)
    - [评估过程](#评估过程)
        - [评估](#评估)
+    - [推理过程](#推理过程)
+        - [导出MindIR](#导出mindir)
+        - [在Ascend310执行推理](#在ascend310执行推理)
+        - [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)
    - [用法](#用法)
        - [推理](#推理)
        - [在预训练模型上继续训练](#在预训练模型上继续训练)
@ -148,10 +153,14 @@ bash scripts/run_eval_ascend.sh $TRAINED_CKPT
    |---README.md    // CNN+CTC相关描述
    |---train.py    // 训练脚本
    |---eval.py    // 评估脚本
+    |---export.py    // 模型导出脚本
+    |---postprocess.py    // 推理后处理脚本
+    |---ascend310_infer    // 用于310推理
    |---scripts
        |---run_standalone_train_ascend.sh    // Ascend单机shell脚本
        |---run_distribute_train_ascend.sh    // Ascend分布式shell脚本
        |---run_eval_ascend.sh    // Ascend评估shell脚本
+        |---run_infer_310.sh    // Ascend310推理的shell脚本
    |---src
        |---__init__.py    // init文件
        |---cnn_ctc.py    // cnn_ctc网络
@ -255,6 +264,39 @@ bash scripts/run_eval_ascend.sh $TRAINED_CKPT

 在IIIT数据集上评估模型，并打印样本结果和总准确率。

+## 推理过程
+
+### 导出MindIR
+
+```shell
+python export.py --ckpt_file [CKPT_PATH] --file_name [FILE_NAME] --file_format [EXPORT_FORMAT]
+```
+
+参数ckpt_file为必填项，
+参数file_name为导出后文件名，
+`EXPORT_FORMAT` 可选 ["AIR", "MINDIR"].
+
+### 在Ascend310执行推理
+
+在执行推理前，mindir文件必须通过`export.py`脚本导出。以下展示了使用mindir模型执行推理的示例。
+目前仅支持batch_size为1的推理，导出模型前请修改`config.py`中的参数`TEST_BATCH_SIZE`为1。
+
+```shell
+# Ascend310 inference
+bash run_infer_310.sh [MINDIR_PATH] [DATA_PATH] [DVPP] [DEVICE_ID]
+```
+
+- `DVPP` 为必填项，需要在["DVPP", "CPU"]选择，大小写均可。CNNCTC目前仅支持使用CPU算子进行推理。
+- `DEVICE_ID` 可选，默认值为0。
+
+### 结果
+
+推理结果保存在脚本执行的当前路径，你可以在acc.log中看到以下精度计算结果。
+
+```bash
+'Accuracy':0.8546
+```
+
 # 模型描述

 ## 性能
@ -290,6 +332,20 @@ bash scripts/run_eval_ascend.sh $TRAINED_CKPT
 | 准确率 | 85% |
 | 推理模型 | 675M（.ckpt文件） |

+### 推理性能
+
+| 参数 | Ascend |
+| -------------- | ---------------------------|
+| 模型版本 | CNNCTC |
+| 资源 | Ascend 310；系统 CentOS 3.10 |
+| 上传日期 | 2021-05-19 |
+| MindSpore版本 | 1.2.0 |
+| 数据集 | IIIT5K |
+| batch_size | 1 |
+| 输出 | Accuracy |
+| 准确率 | Accuracy=0.8546 |
+| 推理模型 | 675M（.ckpt文件） |
+
 ## 用法

 ### 推理
--- a/model_zoo/official/cv/cnnctc/ascend310_infer/CMakeLists.txt
+++ b/model_zoo/official/cv/cnnctc/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)
--- a/model_zoo/official/cv/cnnctc/ascend310_infer/build.sh
+++ b/model_zoo/official/cv/cnnctc/ascend310_infer/build.sh
@ -0,0 +1,23 @@
+#!/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
+  mkdir out
+fi
+cd out || exit
+cmake .. \
+    -DMINDSPORE_PATH="`pip show mindspore-ascend | grep Location | awk '{print $2"/mindspore"}' | xargs realpath`"
+make
--- a/model_zoo/official/cv/cnnctc/ascend310_infer/inc/utils.h
+++ b/model_zoo/official/cv/cnnctc/ascend310_infer/inc/utils.h
@ -0,0 +1,32 @@
+/**
+ * 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);
+#endif
--- a/model_zoo/official/cv/cnnctc/ascend310_infer/src/main.cc
+++ b/model_zoo/official/cv/cnnctc/ascend310_infer/src/main.cc
@ -0,0 +1,198 @@
+/**
+ * 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 <math.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/transforms.h"
+#include "include/dataset/vision_ascend.h"
+#include "include/dataset/execute.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::DataType;
+using mindspore::dataset::Execute;
+using mindspore::dataset::InterpolationMode;
+using mindspore::dataset::TensorTransform;
+using mindspore::dataset::vision::Pad;
+using mindspore::dataset::vision::Resize;
+using mindspore::dataset::vision::HWC2CHW;
+using mindspore::dataset::vision::Normalize;
+using mindspore::dataset::vision::Decode;
+
+
+DEFINE_string(mindir_path, "", "mindir path");
+DEFINE_string(dataset_path, ".", "dataset path");
+DEFINE_int32(device_id, 0, "device id");
+DEFINE_int32(image_height, 32, "image height");
+DEFINE_int32(image_width, 100, "image width");
+
+int PadImage(const MSTensor &input, MSTensor *output) {
+  std::shared_ptr<TensorTransform> normalize(new Normalize({127.5, 127.5, 127.5},
+                                                           {127.5, 127.5, 127.5}));
+  Execute composeNormalize({normalize});
+  std::vector<int64_t> shape = input.Shape();
+  auto imgResize = MSTensor();
+  auto imgNormalize = MSTensor();
+  int paddingSize;
+  int NewWidth;
+  float ratio;
+  ratio = static_cast<float> (shape[1]) / static_cast<float> (shape[0]);
+  NewWidth = ceil(FLAGS_image_height * ratio);
+  paddingSize = FLAGS_image_width - NewWidth;
+  if (NewWidth > FLAGS_image_width) {
+    std::shared_ptr<TensorTransform> resize(new Resize({FLAGS_image_height, FLAGS_image_width},
+                                                       InterpolationMode::kArea));
+    Execute composeResize({resize});
+    composeResize(input, &imgResize);
+    composeNormalize(imgResize, output);
+  } else {
+    std::shared_ptr<TensorTransform> resize(new Resize({FLAGS_image_height, NewWidth},
+                                                       InterpolationMode::kArea));
+    Execute composeResize({resize});
+    composeResize(input, &imgResize);
+    composeNormalize(imgResize, &imgNormalize);
+    std::shared_ptr<TensorTransform> pad(new Pad({0, 0, paddingSize, 0}));
+    Execute composePad({pad});
+    composePad(imgNormalize, output);
+  }
+  return 0;
+}
+
+int main(int argc, char **argv) {
+  gflags::ParseCommandLineFlags(&argc, &argv, true);
+  if (RealPath(FLAGS_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(FLAGS_device_id);
+  ascend310->SetPrecisionMode("allow_fp32_to_fp16");
+  ascend310->SetOpSelectImplMode("high_precision");
+  ascend310->SetBufferOptimizeMode("off_optimize");
+  context->MutableDeviceInfo().push_back(ascend310);
+  mindspore::Graph graph;
+  Serialization::Load(FLAGS_mindir_path, ModelType::kMindIR, &graph);
+  Model model;
+  Status ret = model.Build(GraphCell(graph), context);
+  if (ret != kSuccess) {
+    std::cout << "ERROR: Build failed." << std::endl;
+    return 1;
+  }
+
+  auto all_files = GetAllFiles(FLAGS_dataset_path);
+  std::map<double, double> costTime_map;
+  size_t size = all_files.size();
+  std::shared_ptr<TensorTransform> decode(new Decode());
+  std::shared_ptr<TensorTransform> hwc2chw(new HWC2CHW());
+  Execute composeDecode({decode});
+  Execute composeTranspose({hwc2chw});
+  for (size_t i = 0; i < size; ++i) {
+    struct timeval start = {0};
+    struct timeval end = {0};
+    double startTimeMs;
+    double endTimeMs;
+    std::vector<MSTensor> inputs;
+    std::vector<MSTensor> outputs;
+    auto imgDecode = MSTensor();
+    auto imgTranspose = MSTensor();
+    auto imgPad = MSTensor();
+    auto img = MSTensor();
+    composeDecode(ReadFileToTensor(all_files[i]), &imgDecode);
+    std::vector<int64_t> shape = imgDecode.Shape();
+    float ratio;
+    int NewWidth;
+    ratio = static_cast<float> (shape[1]) / static_cast<float> (shape[0]);
+    NewWidth = ceil(FLAGS_image_height * ratio);
+    PadImage(imgDecode, &imgPad);
+    composeTranspose(imgPad, &img);
+    if (NewWidth < FLAGS_image_width) {
+      int img_size = FLAGS_image_width * FLAGS_image_height * 3;
+      void *netOutput;
+      float *address;
+      netOutput = img.MutableData();
+      address = static_cast<float *>(netOutput);
+      for (int j = 0; j <= img_size; j += FLAGS_image_width) {
+        float temp = 0;
+        address = address + NewWidth - 1;
+        temp = *address;
+        int n = NewWidth;
+        address++;
+        while (n < FLAGS_image_width) {
+          *address = temp;
+          address++;
+          n++;
+        }
+      }
+    }
+
+    std::vector<MSTensor> model_inputs = model.GetInputs();
+    inputs.emplace_back(model_inputs[0].Name(), model_inputs[0].DataType(), model_inputs[0].Shape(),
+                     img.Data().get(), img.DataSize());
+    gettimeofday(&start, nullptr);
+    ret = model.Predict(inputs, &outputs);
+    gettimeofday(&end, nullptr);
+    if (ret != kSuccess) {
+      std::cout << "Predict " << all_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(all_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;
+}
--- a/model_zoo/official/cv/cnnctc/ascend310_infer/src/utils.cc
+++ b/model_zoo/official/cv/cnnctc/ascend310_infer/src/utils.cc
@ -0,0 +1,130 @@
+/**
+ * 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 "inc/utils.h"
+
+#include <fstream>
+#include <algorithm>
+#include <iostream>
+
+using mindspore::MSTensor;
+using mindspore::DataType;
+
+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('.'), ".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;
+}
--- a/model_zoo/official/cv/cnnctc/export.py
+++ b/model_zoo/official/cv/cnnctc/export.py
@ -25,7 +25,7 @@ from src.cnn_ctc import CNNCTC_Model
 parser = argparse.ArgumentParser(description="CNNCTC_export")
 parser.add_argument("--device_id", type=int, default=0, help="Device id")
 parser.add_argument("--file_name", type=str, default="cnn_ctc", help="CNN&CTC output air name.")
-parser.add_argument("--file_format", type=str, choices=["AIR", "ONNX", "MINDIR"], default="AIR", help="file format")
+parser.add_argument("--file_format", type=str, choices=["AIR", "MINDIR"], default="AIR", help="file format")
 parser.add_argument("--device_target", type=str, choices=["Ascend", "GPU", "CPU"], default="Ascend",
                    help="device target")
 parser.add_argument("--ckpt_file", type=str, default="./ckpts/cnn_ctc.ckpt", help="CNN&CTC ckpt file.")
--- a/model_zoo/official/cv/cnnctc/postprocess.py
+++ b/model_zoo/official/cv/cnnctc/postprocess.py
@ -0,0 +1,60 @@
+# 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.
+# ============================================================================
+"""post process for 310 inference"""
+import os
+import argparse
+import numpy as np
+from src.config import Config_CNNCTC
+from src.util import CTCLabelConverter
+
+parser = argparse.ArgumentParser(description="cnnctc acc calculation")
+parser.add_argument("--result_path", type=str, required=True, help="result files path.")
+parser.add_argument("--label_path", type=str, required=True, help="label path.")
+args = parser.parse_args()
+
+
+def calcul_acc(labels, preds):
+    return sum(1 for x, y in zip(labels, preds) if x == y) / len(labels)
+
+
+def get_result(result_path, label_path):
+    config = Config_CNNCTC()
+    converter = CTCLabelConverter(config.CHARACTER)
+    files = os.listdir(result_path)
+    preds = []
+    labels = []
+    label_dict = {}
+    with open(label_path, 'r') as f:
+        lines = f.readlines()
+        for line in lines:
+            label_dict[line.split(',')[0]] = line.split(',')[1].replace('\n', '')
+    for file in files:
+        file_name = file.split('.')[0]
+        label = label_dict[file_name + '.png']
+        labels.append(label)
+        resultPath = os.path.join(result_path, file)
+        output = np.fromfile(resultPath, dtype=np.float32)
+        output = np.reshape(output, (config.FINAL_FEATURE_WIDTH, config.NUM_CLASS))
+        model_predict = np.squeeze(output)
+        preds_size = np.array([model_predict.shape[0]] * 1)
+        preds_index = np.argmax(model_predict, axis=1)
+        preds_str = converter.decode(preds_index, preds_size)
+        preds.append(preds_str[0].upper())
+    acc = calcul_acc(labels, preds)
+    print("TOtal data: {}, accuracy: {}".format(len(labels), acc))
+
+
+if __name__ == '__main__':
+    get_result(args.result_path, args.label_path)
--- a/model_zoo/official/cv/cnnctc/scripts/run_infer_310.sh
+++ b/model_zoo/official/cv/cnnctc/scripts/run_infer_310.sh
@ -0,0 +1,108 @@
+#!/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 || $# -gt 4 ]]; then
+    echo "Usage: bash run_infer_310.sh [MINDIR_PATH] [DATA_PATH] [LABEL_PATH] [DVPP] [DEVICE_ID]
+    DVPP is mandatory, and must choose from [DVPP|CPU], it's case-insensitive
+    DEVICE_ID is optional, it can be set by environment variable device_id, otherwise the value is zero"
+exit 1
+fi
+
+get_real_path(){
+    if [ "${1:0:1}" == "/" ]; then
+        echo "$1"
+    else
+        echo "$(realpath -m $PWD/$1)"
+    fi
+}
+model=$(get_real_path $1)
+data_path=$(get_real_path $2)
+label_path=$(get_real_path $3)
+DVPP=${4^^}
+
+device_id=0
+if [ $# == 5 ]; then
+    device_id=$5
+fi
+
+echo "mindir name: "$model
+echo "dataset path: "$data_path
+echo "label path: "$label_path
+echo "image process mode: "$DVPP
+echo "device id: "$device_id
+
+export ASCEND_HOME=/usr/local/Ascend/
+if [ -d ${ASCEND_HOME}/ascend-toolkit ]; then
+    export PATH=$ASCEND_HOME/ascend-toolkit/latest/fwkacllib/ccec_compiler/bin:$ASCEND_HOME/ascend-toolkit/latest/atc/bin:$PATH
+    export LD_LIBRARY_PATH=/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=${TBE_IMPL_PATH}:$ASCEND_HOME/ascend-toolkit/latest/fwkacllib/python/site-packages:$PYTHONPATH
+    export ASCEND_OPP_PATH=$ASCEND_HOME/ascend-toolkit/latest/opp
+else
+    export PATH=$ASCEND_HOME/atc/ccec_compiler/bin:$ASCEND_HOME/atc/bin:$PATH
+    export LD_LIBRARY_PATH=/usr/local/lib:$ASCEND_HOME/atc/lib64:$ASCEND_HOME/acllib/lib64:$ASCEND_HOME/driver/lib64:$ASCEND_HOME/add-ons:$LD_LIBRARY_PATH
+    export PYTHONPATH=$ASCEND_HOME/atc/python/site-packages:$PYTHONPATH
+    export ASCEND_OPP_PATH=$ASCEND_HOME/opp
+fi
+
+function compile_app()
+{
+    cd ../ascend310_infer || exit
+    bash build.sh &> build.log
+}
+
+function infer()
+{
+    cd - || exit
+    if [ -d result_Files ]; then
+        rm -rf ./result_Files
+    fi
+    if [ -d time_Result ]; then
+        rm -rf ./time_Result
+    fi
+    mkdir result_Files
+    mkdir time_Result
+    if [ "$DVPP" == "DVPP" ];then
+      echo "CNNCTC ONLY SUPPORT CPU MODE"
+      exit 1
+    elif [ "$DVPP" == "CPU"  ]; then
+      ../ascend310_infer/out/main --mindir_path=$model --dataset_path=$data_path --device_id=$device_id --image_height=32 --image_width=100 &> infer.log
+    else
+      echo "image process mode must be in [DVPP|CPU]"
+      exit 1
+    fi
+}
+
+function cal_acc()
+{
+    python3.7 ../postprocess.py --result_path=./result_Files --label_path=$label_path &> acc.log &
+}
+
+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
--- a/model_zoo/official/cv/lenet/README.md
+++ b/model_zoo/official/cv/lenet/README.md
@ -179,8 +179,8 @@ The ckpt_file parameter is required,

 ### Infer on Ascend310

-Before performing inference, the mindir file must bu exported by `export.py` script. We only provide an example of inference using MINDIR model.
-Current batch_Size can only be set to 1.
+Before performing inference, the mindir file must be exported by `export.py` script. We only provide an example of inference using MINDIR model.
+Current batch_size can only be set to 1.

 ```shell
 # Ascend310 inference
--- a/model_zoo/official/cv/lenet/README_CN.md
+++ b/model_zoo/official/cv/lenet/README_CN.md
@ -181,8 +181,8 @@ python export.py --ckpt_file [CKPT_PATH] --file_name [FILE_NAME] --file_format [

 ### 在Ascend310执行推理

-在执行推理前，mindir文件必须通过`export.py`脚本导出。以下展示了使用minir模型执行推理的示例。
-目前仅支持batch_Size为1的推理。
+在执行推理前，mindir文件必须通过`export.py`脚本导出。以下展示了使用mindir模型执行推理的示例。
+目前仅支持batch_size为1的推理。

 ```shell
 # Ascend310 inference
--- a/model_zoo/official/cv/mobilenetv2/README.md
+++ b/model_zoo/official/cv/mobilenetv2/README.md
@ -247,8 +247,8 @@ The ckpt_file parameter is required,

 ### Infer on Ascend310

-Before performing inference, the mindir file must bu exported by `export.py` script. We only provide an example of inference using MINDIR model.
-Current batch_Size can only be set to 1.
+Before performing inference, the mindir file must be exported by `export.py` script. We only provide an example of inference using MINDIR model.
+Current batch_size can only be set to 1.

 ```shell
 # Ascend310 inference
--- a/model_zoo/official/cv/mobilenetv2/README_CN.md
+++ b/model_zoo/official/cv/mobilenetv2/README_CN.md
@ -254,8 +254,8 @@ python export.py --platform [PLATFORM] --ckpt_file [CKPT_PATH] --file_format [EX

 ### 在Ascend310执行推理

-在执行推理前，mindir文件必须通过`export.py`脚本导出。以下展示了使用minir模型执行推理的示例。
-目前仅支持batch_Size为1的推理。
+在执行推理前，mindir文件必须通过`export.py`脚本导出。以下展示了使用mindir模型执行推理的示例。
+目前仅支持batch_size为1的推理。

 ```shell
 # Ascend310 inference
--- a/model_zoo/official/cv/warpctc/README.md
+++ b/model_zoo/official/cv/warpctc/README.md
@ -16,10 +16,15 @@
        - [Distributed Training](#distributed-training)
    - [Evaluation Process](#evaluation-process)
        - [Evaluation](#evaluation)
+    - [Inference Process](#inference-process)
+        - [Export MindIR](#export-mindir)
+        - [Infer on Ascend310](#infer-on-ascend310)
+        - [result](#result)
 - [Model Description](#model-description)
    - [Performance](#performance)
        - [Training Performance](#training-performance)
        - [Evaluation Performance](#evaluation-performance)
+        - [Inference Performance](#inference-performance)
 - [Description of Random Situation](#description-of-random-situation)
 - [ModelZoo Homepage](#modelzoo-homepage)

@ -167,10 +172,12 @@ The dataset is self-generated using a third-party library called [captcha](https
 └──warpctc
  ├── README.md                         # descriptions of warpctc
  ├── README_CN.md                      # chinese descriptions of warpctc
+  ├── ascend310_infer                   # application for 310 inference
  ├── script
    ├── run_distribute_train.sh         # launch distributed training in Ascend(8 pcs)
    ├── run_distribute_train_for_gpu.sh # launch distributed training in GPU
    ├── run_eval.sh                     # launch evaluation
+    ├── run_infer_310.sh                # launch 310infer
    ├── run_process_data.sh             # launch dataset generation
    └── run_standalone_train.sh         # launch standalone training(1 pcs)
  ├── src
@ -190,6 +197,8 @@ The dataset is self-generated using a third-party library called [captcha](https
  ├── mindspore_hub_conf.py             # mindspore hub interface
  ├── eval.py                           # eval net
  ├── process_data.py                   # dataset generation script
+  ├── postprocess.py                    # 310infer postprocess script
+  ├── preprocess.py                     # 310infer preprocess script
  └── train.py                          # train net
 ```

@ -267,6 +276,39 @@ bash run_distribute_train_gpu.sh [RANK_SIZE] [TRAIN_DATA_DIR]
 bash run_eval.sh [TEST_DATA_DIR] [CHECKPOINT_PATH] [DEVICE_TARGET]
 ```

+## [Inference Process](#contents)
+
+### Export MindIR
+
+```shell
+python export.py --ckpt_file [CKPT_PATH] --file_name [FILE_NAME] --file_format [FILE_FORMAT]
+```
+
+The ckpt_file parameter is required,
+`EXPORT_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.
+Current batch_size can only be set to 1.
+Use mindir+bin method for inferring, and bin is a binary format file of the preprocessed picture.
+
+```shell
+# Ascend310 inference
+bash run_infer_310.sh [MINDIR_PATH] [DATA_PATH] [DEVICE_ID]
+```
+
+- `DATA_PATH` is mandatory,the data format is the path of the bin.
+- `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.
+
+```bash
+'Accuracy': 0.952
+```
+
 ## [Model Description](#contents)

 ### [Performance](#contents)
@ -305,6 +347,20 @@ bash run_eval.sh [TEST_DATA_DIR] [CHECKPOINT_PATH] [DEVICE_TARGET]
 | Accuracy            | 99.0%                       |
 | Model for inference | 20.3M (.ckpt file)          |

+#### Inference Performance
+
+| Parameters          | Ascend                      |
+| ------------------- | --------------------------- |
+| Model Version       | WarpCTC                     |
+| Resource            | Ascend 310; CentOS 3.10     |
+| Uploaded Date       | 24/05/2021 (month/day/year) |
+| MindSpore Version   | 1.2.0                       |
+| Dataset             | Captcha                     |
+| batch_size          | 1                           |
+| outputs             | Accuracy                    |
+| Accuracy            | Accuracy=0.952              |
+| Model for inference | 40.6M(.ckpt file)           |
+
 ## [Description of Random Situation](#contents)

 In dataset.py, we set the seed inside “create_dataset" function. We also use random seed in train.py for weight initialization.
--- a/model_zoo/official/cv/warpctc/README_CN.md
+++ b/model_zoo/official/cv/warpctc/README_CN.md
@ -19,10 +19,15 @@
        - [分布式训练](#分布式训练)
    - [评估过程](#评估过程)
        - [评估](#评估)
+    - [推理过程](#推理过程)
+        - [导出MindIR](#导出mindir)
+        - [在Ascend310执行推理](#在ascend310执行推理)
+        - [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)
 - [随机情况说明](#随机情况说明)
 - [ModelZoo主页](#modelzoo主页)

@ -171,10 +176,12 @@ WarpCTC是带有一层FC神经网络的二层堆叠LSTM模型。详细信息请
 └──warpctc
  ├── README.md                         # warpctc文档说明
  ├── README_CN.md                      # warpctc中文文档说明
+  ├── ascend310_infer                   # 用于310推理
  ├── script
    ├── run_distribute_train.sh         # 启动Ascend分布式训练（8卡）
    ├── run_distribute_train_for_gpu.sh # 启动GPU分布式训练
    ├── run_eval.sh                     # 启动评估
+    ├── run_infer_310.sh                # 启动310推理
    ├── run_process_data.sh             # 启动数据集生成
    └── run_standalone_train.sh         # 启动单机训练（1卡）
  ├── src
@ -194,6 +201,8 @@ WarpCTC是带有一层FC神经网络的二层堆叠LSTM模型。详细信息请
  ├── mindspore_hub_conf.py             # Mindspore Hub接口
  ├── eval.py                           # 评估网络
  ├── process_data.py                   # 数据集生成脚本
+  ├── postprocess.py                    # 310推理后处理脚本
+  ├── preprocess.py                     # 将数据前处理脚本
  └── train.py                          # 训练网络
 ```

@ -271,6 +280,39 @@ bash run_distribute_train_gpu.sh [RANK_SIZE] [TRAIN_DATA_DIR]
 bash run_eval.sh [TEST_DATA_DIR] [CHECKPOINT_PATH] [DEVICE_TARGET]
 ```

+## 推理过程
+
+### 导出MindIR
+
+```shell
+python export.py --ckpt_file [CKPT_PATH] --file_name [FILE_NAME] --file_format [FILE_FORMAT]
+```
+
+参数ckpt_file为必填项，
+`EXPORT_FORMAT` 必须在 ["AIR", "MINDIR"]中选择。
+
+### 在Ascend310执行推理
+
+在执行推理前，mindir文件必须通过`export.py`脚本导出。以下展示了使用mindir模型执行推理的示例。
+目前仅支持batch_size为1的推理。
+采用mindir+bin方式进行推理，其中bin为预处理完的图片的二进制文件。
+
+```shell
+# Ascend310 inference
+bash run_infer_310.sh [MINDIR_PATH] [DATA_PATH] [DEVICE_ID]
+```
+
+- `DATA_PATH` 为必填项，数据格式为bin的路径。
+- `DEVICE_ID` 可选，默认值为0。
+
+### 结果
+
+推理结果保存在脚本执行的当前路径，你可以在acc.log中看到以下精度计算结果。
+
+```bash
+'Accuracy':0.952
+```
+
 ## 模型描述

 ### 性能
@ -309,6 +351,20 @@ bash run_eval.sh [TEST_DATA_DIR] [CHECKPOINT_PATH] [DEVICE_TARGET]
 | 准确率            | 99.0%                       |
 | 推理模型 | 20.3M (.ckpt文件)          |

+### 推理性能
+
+| 参数            | Ascend                      |
+| ------------- | ----------------------------|
+| 模型版本        | WarpCTC                     |
+| 资源           | Ascend 310；系统 CentOS 3.10 |
+| 上传日期        | 2021-05-24                  |
+| MindSpore版本  | 1.2.0                       |
+| 数据集          | Captcha                     |
+| batch_size     | 1                          |
+| 输出            | Accuracy                   |
+| 准确率          | Accuracy=0.952             |
+| 推理模型        | 40.6M（.ckpt文件）           |
+
 ## 随机情况说明

 在dataset.py中设置“create_dataset”函数内的种子。使用train.py中的随机种子进行权重初始化。
--- a/model_zoo/official/cv/warpctc/ascend310_infer/CMakeLists.txt
+++ b/model_zoo/official/cv/warpctc/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)
--- a/model_zoo/official/cv/warpctc/ascend310_infer/build.sh
+++ b/model_zoo/official/cv/warpctc/ascend310_infer/build.sh
@ -0,0 +1,23 @@
+#!/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
+  mkdir out
+fi
+cd out || exit
+cmake .. \
+    -DMINDSPORE_PATH="`pip show mindspore-ascend | grep Location | awk '{print $2"/mindspore"}' | xargs realpath`"
+make
--- a/model_zoo/official/cv/warpctc/ascend310_infer/inc/utils.h
+++ b/model_zoo/official/cv/warpctc/ascend310_infer/inc/utils.h
@ -0,0 +1,32 @@
+/**
+ * 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);
+#endif
--- a/model_zoo/official/cv/warpctc/ascend310_infer/src/main.cc
+++ b/model_zoo/official/cv/warpctc/ascend310_infer/src/main.cc
@ -0,0 +1,127 @@
+/**
+ * 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 "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;
+
+DEFINE_string(mindir_path, "", "mindir path");
+DEFINE_string(input0_path, ".", "input0 path");
+DEFINE_int32(device_id, 0, "device id");
+
+int main(int argc, char **argv) {
+  gflags::ParseCommandLineFlags(&argc, &argv, true);
+  if (RealPath(FLAGS_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(FLAGS_device_id);
+  context->MutableDeviceInfo().push_back(ascend310);
+  mindspore::Graph graph;
+  Serialization::Load(FLAGS_mindir_path, ModelType::kMindIR, &graph);
+
+  Model model;
+  Status ret = model.Build(GraphCell(graph), context);
+  if (ret != kSuccess) {
+    std::cout << "ERROR: Build failed." << std::endl;
+    return 1;
+  }
+
+  std::vector<MSTensor> model_inputs = model.GetInputs();
+  if (model_inputs.empty()) {
+    std::cout << "Invalid model, inputs is empty." << std::endl;
+    return 1;
+  }
+
+  auto input0_files = GetAllFiles(FLAGS_input0_path);
+
+  if (input0_files.empty()) {
+    std::cout << "ERROR: input data empty." << std::endl;
+    return 1;
+  }
+
+  std::map<double, double> costTime_map;
+  size_t size = input0_files.size();
+
+  for (size_t i = 0; i < size; ++i) {
+    struct timeval start = {0};
+    struct timeval end = {0};
+    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);
+    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;
+}
--- a/model_zoo/official/cv/warpctc/ascend310_infer/src/utils.cc
+++ b/model_zoo/official/cv/warpctc/ascend310_infer/src/utils.cc
@ -0,0 +1,130 @@
+/**
+ * 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 "inc/utils.h"
+
+#include <fstream>
+#include <algorithm>
+#include <iostream>
+
+using mindspore::MSTensor;
+using mindspore::DataType;
+
+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('.'), ".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;
+}
--- a/model_zoo/official/cv/warpctc/postprocess.py
+++ b/model_zoo/official/cv/warpctc/postprocess.py
@ -0,0 +1,84 @@
+# 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.
+# ============================================================================
+"""post process for 310 inference"""
+import os
+import argparse
+import numpy as np
+
+batch_Size = 1
+parser = argparse.ArgumentParser(description="warpctc acc calculation")
+parser.add_argument("--result_path", type=str, required=True, help="result files path.")
+parser.add_argument("--label_path", type=str, required=True, help="label path.")
+args = parser.parse_args()
+
+
+def is_eq(pred_lbl, target):
+    pred_diff = len(target) - len(pred_lbl)
+    if pred_diff > 0:
+        pred_lbl.extend([10] * pred_diff)
+    return pred_lbl == target
+
+
+def get_prediction(y_pred):
+    seq_len, batch_size, _ = y_pred.shape
+    indices = y_pred.argmax(axis=2)
+    lens = [seq_len] * batch_size
+    pred_lbl = []
+    for i in range(batch_size):
+        idx = indices[:, i]
+        last_idx = 10
+        pred_lbl = []
+        for j in range(lens[i]):
+            cur_idx = idx[j]
+            if cur_idx not in [last_idx, 10]:
+                pred_lbl.append(cur_idx)
+            last_idx = cur_idx
+    return pred_lbl
+
+
+def calcul_acc(y_pred, y):
+    correct_num = 0
+    total_num = 0
+    for b_idx, target in enumerate(y):
+        if is_eq(y_pred[b_idx], target):
+            correct_num += 1
+        total_num += 1
+    if total_num == 0:
+        raise RuntimeError('Accuracy can not be calculated, because the number of samples is 0.')
+    return correct_num / total_num
+
+
+def get_result(result_path, label_path):
+    files = os.listdir(result_path)
+    preds = []
+    labels = []
+    label_dict = {}
+    with open(label_path, 'r') as file:
+        lines = file.readlines()
+        for line in lines:
+            label_dict[line.split(',')[0]] = np.array(
+                line.replace('\n', '').replace('[', '').replace(']', '').split(',')[1:]).astype(dtype=int).tolist()
+    for file in files:
+        label = label_dict[file]
+        labels.append(label)
+        resultPath = os.path.join(result_path, file)
+        output = np.fromfile(resultPath, dtype=np.float16).reshape((-1, batch_Size, 11))
+        preds.append(get_prediction(output))
+    acc = calcul_acc(preds, labels)
+    print("Total data: {}, accuracy: {}".format(len(labels), acc))
+
+
+if __name__ == '__main__':
+    get_result(args.result_path, args.label_path)
--- a/model_zoo/official/cv/warpctc/preprocess.py
+++ b/model_zoo/official/cv/warpctc/preprocess.py
@ -0,0 +1,49 @@
+# Copyright 2020-2021 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.
+# ============================================================================
+import os
+import math as m
+import argparse
+from src.config import config as cf
+from src.dataset import create_dataset
+
+batch_size = 1
+parser = argparse.ArgumentParser(description="Warpctc preprocess")
+parser.add_argument("--dataset_path", type=str, default=None, help="Dataset, default is None.")
+parser.add_argument("--output_path", type=str, default=None, help="output path")
+
+args_opt = parser.parse_args()
+
+if __name__ == "__main__":
+    input_size = m.ceil(cf.captcha_height / 64) * 64 * 3
+    dataset = create_dataset(dataset_path=args_opt.dataset_path,
+                             batch_size=batch_size,
+                             device_target="Ascend")
+
+    img_path = args_opt.output_path
+    if not os.path.isdir(img_path):
+        os.makedirs(img_path)
+    total = dataset.get_dataset_size()
+    iter_num = 0
+    label_dict = {}
+    for data in dataset.create_dict_iterator(output_numpy=True):
+        file_name = str(iter_num) + ".bin"
+        img = data["image"]
+        label_dict[file_name] = data["label"][0].tolist()
+        img.tofile(os.path.join(img_path, file_name))
+        iter_num += 1
+    with open('./label.txt', 'w') as file:
+        for k, v in label_dict.items():
+            file.write(str(k) + ',' + str(v) + '\n')
+    print("total image num:", total)
--- a/model_zoo/official/cv/warpctc/scripts/run_infer_310.sh
+++ b/model_zoo/official/cv/warpctc/scripts/run_infer_310.sh
@ -0,0 +1,109 @@
+#!/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 || $# -gt 3 ]]; then
+    echo "Usage: bash run_infer_310.sh [MINDIR_PATH] [DATA_PATH] [DEVICE_ID]"
+exit 1
+fi
+
+get_real_path(){
+    if [ "${1:0:1}" == "/" ]; then
+        echo "$1"
+    else
+        echo "$(realpath -m $PWD/$1)"
+    fi
+}
+model=$(get_real_path $1)
+input0_path=$(get_real_path $2)
+
+device_id=0
+if [ $# == 3 ]; then
+    device_id=$3
+fi
+
+echo "mindir name: "$model
+echo "input0 path: "$input0_path
+echo "device id: "$device_id
+
+export ASCEND_HOME=/usr/local/Ascend/
+if [ -d ${ASCEND_HOME}/ascend-toolkit ]; then
+    export PATH=$ASCEND_HOME/ascend-toolkit/latest/fwkacllib/ccec_compiler/bin:$ASCEND_HOME/ascend-toolkit/latest/atc/bin:$PATH
+    export LD_LIBRARY_PATH=/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=${TBE_IMPL_PATH}:$ASCEND_HOME/ascend-toolkit/latest/fwkacllib/python/site-packages:$PYTHONPATH
+    export ASCEND_OPP_PATH=$ASCEND_HOME/ascend-toolkit/latest/opp
+else
+    export PATH=$ASCEND_HOME/atc/ccec_compiler/bin:$ASCEND_HOME/atc/bin:$PATH
+    export LD_LIBRARY_PATH=/usr/local/lib:$ASCEND_HOME/atc/lib64:$ASCEND_HOME/acllib/lib64:$ASCEND_HOME/driver/lib64:$ASCEND_HOME/add-ons:$LD_LIBRARY_PATH
+    export PYTHONPATH=$ASCEND_HOME/atc/python/site-packages:$PYTHONPATH
+    export ASCEND_OPP_PATH=$ASCEND_HOME/opp
+fi
+
+function preprocess_data()
+{
+    if [ -d preprocess_Result ]; then
+        rm -rf ./preprocess_Result
+    fi
+    mkdir preprocess_Result
+    python3.7 ../preprocess.py --dataset_path=$input0_path --output_path=./preprocess_Result &> preprocess.log &
+    input0_path=./preprocess_Result
+}
+
+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=$model --input0_path=$input0_path --device_id=$device_id &> infer.log
+}
+
+function cal_acc()
+{
+    python3.7 ../postprocess.py --result_path=./result_Files --label_path=./label.txt &> acc.log &
+}
+
+preprocess_data
+if [ $? -ne 0 ]; then
+    echo "preprocess data failed"
+    exit 1
+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