p83571920
/
mindspore
forked from mindspore-Ecosystem/mindspore
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

WideResNet

This commit is contained in:
yangwm 2021-09-06 17:57:00 +08:00
parent 5f92471380
commit 29f0c48e3b
11 changed files with 664 additions and 45 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

98
model_zoo/research/cv/wideresnet/README_CN.md
View File

@ -15,10 +15,14 @@
- [用法](#用法)
- [Ascend处理器环境运行](#ascend处理器环境运行)
- [结果](#结果)
- [评估过程](#评估过程)
- [用法](#用法)
- [Ascend处理器环境运行](#ascend处理器环境运行)
- [结果](#结果)
- [评估过程](#评估过程)
- [用法](#用法)
- [Ascend处理器环境运行](#ascend处理器环境运行)
- [结果](#结果)
- [Ascend310推理过程](#推理过程)
- [导出MindIR](#导出MindIR)
- [在Acsend310执行推理](#在Acsend310执行推理)
- [结果](#结果)
- [模型描述](#模型描述)
- [性能](#性能)
- [评估性能](#评估性能)
@ -55,15 +59,13 @@ WideResNet的总体网络架构如下：[链接](https://arxiv.org/abs/1605.0714
- 下载数据集，目录结构如下：
```text
└─train
└─cifar-10-batches-bin
├─data_batch_1.bin # 训练数据集
├─data_batch_2.bin # 训练数据集
├─data_batch_3.bin # 训练数据集
├─data_batch_4.bin # 训练数据集
├─data_batch_5.bin # 训练数据集
└─test_batch.bin # 评估数据集
└─eval
└─test_batch.bin # 评估数据集
```
# 环境要求
@ -84,33 +86,34 @@ WideResNet的总体网络架构如下：[链接](https://arxiv.org/abs/1605.0714
```Shell
# 分布式训练
用法：
cd scripts
bash run_distribute_train.sh [RANK_TABLE_FILE] [DATASET_PATH] [PRETRAINED_CKPT_PATH] [MODELART]
用法：bash run_distribute_train.sh [RANK_TABLE_FILE] [DATASET_PATH] [PRETRAINED_CKPT_PATH]（可选）
# 单机训练
用法：
cd scripts
bash run_standalone_train.sh [DATASET_PATH] [PRETRAINED_CKPT_PATH] [MODELART]
用法：bash run_standalone_train.sh [DATASET_PATH] [PRETRAINED_CKPT_PATH]（可选）
# 运行评估示例
用法：
cd scripts
bash run_eval.sh [DATASET_PATH] [CHECKPOINT_PATH] [MODELART]
用法：bash run_eval.sh [DATASET_PATH] [CHECKPOINT_PATH]
```
若没有[PRETRAINED_CKPT_PATH]，使用 “” 作为参数运行脚本。
# 脚本说明
## 脚本及样例代码
```text
└──wideresnet
├── README.md
├── README_CN.md
├── ascend310_infer
├── inc
├── util.h
├── src
├── build.sh
├── CMakeList.txt
├── main.cc
├── utils.cc
├── scripts
├── run_distribute_train.sh # 启动Ascend分布式训练（8卡）
├── run_eval.sh # 启动Ascend评估
├── run_eval.sh # 启动Ascend910评估
├── run_infer_310.sh # 启动Ascend310评估
└── run_standalone_train.sh # 启动Ascend单机训练（单卡）
├── src
├── config.py # 参数配置
@ -119,9 +122,11 @@ bash run_eval.sh [DATASET_PATH] [CHECKPOINT_PATH] [MODELART]
├── generator_lr.py # 生成每个步骤的学习率
├── save_callback.py # 自定义回调函数保存最优ckpt
└── wide_resnet.py # WideResNet网络结构
├── eval.py # 评估网络
├── export.py # 导出网络
└── train.py # 训练网络
├── eval.py # 910评估网络
├── export.py # 910导出网络
├── postprocess.py # 310推理精度计算
├── preprocess.py # 310推理前数据处理
└── train.py # 910训练网络
```
# 脚本参数
@ -159,19 +164,13 @@ bash run_eval.sh [DATASET_PATH] [CHECKPOINT_PATH] [MODELART]
```Shell
# 分布式训练
用法：
cd scripts
bash run_distribute_train.sh [RANK_TABLE_FILE] [DATASET_PATH] [PRETRAINED_CKPT_PATH] [MODELART]
用法：bash run_distribute_train.sh [RANK_TABLE_FILE] [DATASET_PATH] [PRETRAINED_CKPT_PATH]（可选）
# 单机训练
用法：
cd scripts
bash run_standalone_train.sh [DATASET_PATH] [PRETRAINED_CKPT_PATH] [MODELART]
用法：bash run_standalone_train.sh [DATASET_PATH] [PRETRAINED_CKPT_PATH]（可选）
```
若没有[PRETRAINED_CKPT_PATH]，使用 “” 作为参数运行脚本。
分布式训练需要提前创建JSON格式的HCCL配置文件。
具体操作，参见[hccn_tools](https://gitee.com/mindspore/mindspore/tree/master/model_zoo/utils/hccl_tools)中的说明。
@ -219,16 +218,12 @@ epoch: 4 step: 195, loss is 1.221174
```Shell
# 评估
用法：
cd scripts
bash run_eval.sh [DATASET_PATH] [CHECKPOINT_PATH] [MODELART]
Usage: bash run_eval.sh [DATASET_PATH] [CHECKPOINT_PATH]
```
```Shell
# 评估示例
用法：
cd scripts
bash run_eval.sh /cifar10 WideResNet_best.ckpt
bash run_eval.sh /cifar10 WideResNet_best.ckpt
```
训练过程中可以生成检查点。
@ -243,6 +238,35 @@ bash run_eval.sh /cifar10 WideResNet_best.ckpt
result: {'top_1_accuracy': 0.9622395833333334}
```
# Ascend310推理过程
## 导出MindIR
```shell
python export.py --ckpt_file [CKPT_PATH] --file_format [FILE_FORMAT] --device_id [0]
```
参数ckpt_file为必填项，
`file_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。
```shell
# Ascend310 inference
bash run_infer_310.sh [MINDIR_PATH] [DATASET_PATH] [DEVICE_ID]
```
- `MINDIR_PATH` mindir文件路径
- `DATASET_PATH` 推理数据集路径
- `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，
你可以在当前文件夹中acc.log查看推理精度，在time_Result中查看推理时间。
# 模型描述
## 性能

33
model_zoo/research/cv/wideresnet/ascend310_infer/inc/utils.h Normal file
View File

@ -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

14
model_zoo/research/cv/wideresnet/ascend310_infer/src/CMakeLists.txt Normal file
View File

@ -0,0 +1,14 @@
cmake_minimum_required(VERSION 3.14.1)
project(MindSporeCxxTestcase[CXX])
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 main.cc utils.cc)
target_link_libraries(main ${MS_LIB} ${MD_LIB} gflags)

18
model_zoo/research/cv/wideresnet/ascend310_infer/src/build.sh Normal file
View File

@ -0,0 +1,18 @@
#!/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.
# ============================================================================
cmake . -DMINDSPORE_PATH="`pip3.7 show mindspore-ascend | grep Location | awk '{print $2"/mindspore"}' | xargs realpath`"
make

142
model_zoo/research/cv/wideresnet/ascend310_infer/src/main.cc Normal file
View File

@ -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::dataset::vision::Decode;
using mindspore::dataset::vision::Resize;
using mindspore::dataset::vision::CenterCrop;
using mindspore::dataset::vision::Normalize;
using mindspore::dataset::vision::HWC2CHW;
using mindspore::dataset::TensorTransform;
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;
DEFINE_string(mindir_path, "", "mindir path");
DEFINE_string(dataset_path, ".", "dataset 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;
}
auto all_files = GetAllFiles(FLAGS_dataset_path);
if (all_files.empty()) {
std::cout << "ERROR: no input data." << std::endl;
return 1;
}
std::vector<MSTensor> modelInputs = model.GetInputs();
std::map<double, double> costTime_map;
size_t size = all_files.size();
std::shared_ptr<TensorTransform> hwc2chw = std::make_shared<HWC2CHW>();
std::shared_ptr<TensorTransform> normalize = std::make_shared<Normalize>(
std::vector<float>{0.4914, 0.4822, 0.4465}, std::vector<float>{0.2023, 0.1994, 0.2010});
std::vector<std::shared_ptr<TensorTransform>> trans_list;
trans_list = {normalize, hwc2chw};
mindspore::dataset::Execute SingleOp(trans_list);
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:" << all_files[i] <<std::endl;
MSTensor image = ReadFileToTensor(all_files[i]);
SingleOp(image, &image);
inputs.emplace_back(modelInputs[0].Name(), modelInputs[0].DataType(), modelInputs[0].Shape(),
image.Data().get(), image.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++) {
average += iter->second - iter->first;
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;
}

145
model_zoo/research/cv/wideresnet/ascend310_infer/src/utils.cc Normal file
View File

@ -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;
}

10
model_zoo/research/cv/wideresnet/export.py
View File

@ -26,6 +26,7 @@ from mindspore import Tensor, load_checkpoint, load_param_into_net, export, cont
from src.wide_resnet import wideresnet
parser = argparse.ArgumentParser(description='WideResNet export')
parser.add_argument("--device_id", type=int, default=0, help="Device id")
parser.add_argument("--run_modelart", type=ast.literal_eval, default=False, help="Run on modelArt, default is false.")
parser.add_argument('--data_url', default=None, help='Directory contains cifar10 dataset.')
parser.add_argument('--train_url', default=None, help='Directory contains checkpoint file')
@ -34,8 +35,9 @@ parser.add_argument("--batch_size", type=int, default=1, help="batch size")
parser.add_argument('--file_format', type=str, choices=["AIR", "ONNX", "MINDIR"], default='AIR', help='file format')
args = parser.parse_args()
context.set_context(mode=context.GRAPH_MODE, device_target="Ascend")
context.set_context(device_id=int(os.environ["DEVICE_ID"]))
context.set_context(mode=context.GRAPH_MODE, device_target=args.device_target)
if args.device_target == "Ascend":
context.set_context(device_id=args.device_id)
if args.run_modelart:
import moxing as mox
@ -48,10 +50,10 @@ if __name__ == '__main__':
net = wideresnet()
param_dict = load_checkpoint(os.path.join(local_output_url, args.ckpt_file))
print('load ckpt')
load_param_into_net(net, param_dict)
param_dict = load_checkpoint(args.ckpt_file)
print('load ckpt to net')
load_param_into_net(net, param_dict)
net.set_train(False)
input_arr = Tensor(np.ones([args.batch_size, 3, 32, 32]), mstype.float32)
print('input')

72
model_zoo/research/cv/wideresnet/postprocess.py Normal file
View File

@ -0,0 +1,72 @@
# 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="WideResNet inference")
parser.add_argument("--result_path", required=True, help="result files path.")
parser.add_argument("--label_path", required=True, help="image file path.")
args = parser.parse_args()
def get_top5_acc(top5_arg, gt_class):
sub_count = 0
for top5, gt in zip(top5_arg, gt_class):
if gt in top5:
sub_count += 1
return sub_count
def cal_acc_cifar10(result_path, label_path):
"""
result_path: path of preprocess image
label_path: path of label
"""
top1_correct = 0
top5_correct = 0
img_tot = 0
result_shape = (1, 10)
files = os.listdir(result_path)
for file in files:
full_file_path = os.path.join(result_path, file)
if os.path.isfile(full_file_path):
result = np.fromfile(full_file_path, dtype=np.float32).reshape(result_shape)
label_file = os.path.join(label_path, file.split(".bin")[0][:-2] + ".bin")
gt_classes = np.fromfile(label_file, dtype=np.int32)
top1_output = np.argmax(result, (-1))
top5_output = np.argsort(result)[:, -5:]
t1_correct = np.equal(top1_output, gt_classes).sum()
top1_correct += t1_correct
top5_correct += get_top5_acc(top5_output, [gt_classes])
img_tot += 1
top1_correct = float(top1_correct)
img_tot = float(img_tot)
top1_acc = top1_correct/img_tot
print("top1_acc", top1_acc)
top5_correct = float(top5_correct)
top5_acc = top5_correct/img_tot
print("top5_acc", top5_acc)
if __name__ == '__main__':
cal_acc_cifar10(args.result_path, args.label_path)

49
model_zoo/research/cv/wideresnet/preprocess.py Normal file
View File

@ -0,0 +1,49 @@
# 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.
# ============================================================================
"""train WideResNet."""
import os
import argparse
from src.dataset import create_dataset
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Ascend WideResnet cifar10 310 preprocess')
parser.add_argument('--data_path', type=str, required=True, help='Location of data')
parser.add_argument('--output_path', type=str, required=True, help='Location of output data.')
parser.add_argument('--device_id', required=True, default=0, help='device_id')
args = parser.parse_args()
# create dataset
dataset = create_dataset(dataset_path=args.data_path, do_train=False,
infer_910=False, device_id=args.device_id, batch_size=1)
step_size = dataset.get_dataset_size()
img_path = os.path.join(args.output_path, "img_data")
label_path = os.path.join(args.output_path, "label")
os.makedirs(img_path)
os.makedirs(label_path)
for idx, data in enumerate(dataset.create_dict_iterator(output_numpy=True, num_epochs=1)):
img_data = data["image"]
img_label = data["label"]
file_name = "google_cifar10_1_" + str(idx) + ".bin"
img_file_path = os.path.join(img_path, file_name)
img_data.tofile(img_file_path)
label_file_path = os.path.join(label_path, file_name)
img_label.tofile(label_file_path)
print("=" * 20, "export bin files finished", "=" * 20)

116
model_zoo/research/cv/wideresnet/scripts/run_infer_310.sh Normal file
View File

@ -0,0 +1,116 @@
#!/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 || $# -gt 3 ]]; then
echo "Usage: sh run_infer_310.sh [MINDIR_PATH] [DATA_PATH] [DEVICE_ID]
DEVICE_ID is optional, it can be set by environment variable device_id, otherwise the value is zero"
exit 1
fi
get_real_path(){
if [ "${1:0:1}" == "/" ]; then
echo "$1"
else
echo "$(realpath -m $PWD/$1)"
fi
}
model=$(get_real_path $1)
data_path=$(get_real_path $2)
device_id=0
if [ $# == 3 ]; then
device_id=$3
fi
echo "mindir name: "$model
echo "dataset path: "$data_path
echo "device id: "$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=$ASCEND_HOME/opp
fi
function compile_app()
{
cd ../ascend310_infer/src/ || exit
if [ -f "Makefile" ]; then
make clean
fi
sh build.sh &> build.log
}
function preprocess_data()
{
if [ -d preprocess_Result ]; then
rm -rf ./preprocess_Result
fi
mkdir preprocess_Result
python3.7 ../preprocess.py --data_path=$data_path --output_path=./preprocess_Result --device_id=$device_id &> preprocess.log
}
function infer()
{
cd - || exit
if [ -d result_Files ]; then
rm -rf ./result_Files
fi
if [ -d time_Result ]; then
rm -rf ./time_Result
fi
mkdir result_Files
mkdir time_Result
../ascend310_infer/src/main --mindir_path=$model --dataset_path=$data_path --device_id=$device_id &> infer.log
}
function cal_acc()
{
python ../postprocess.py --label_path=./preprocess_Result/label --result_path=result_Files &> acc.log
if [ $? -ne 0 ]; then
echo "calculate accuracy failed"
exit 1
fi
}
preprocess_data
data_path=./preprocess_Result/img_data
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

12
model_zoo/research/cv/wideresnet/src/dataset.py
View File

@ -23,7 +23,7 @@ import mindspore.dataset.vision.c_transforms as C
import mindspore.dataset.transforms.c_transforms as C2
def create_dataset(dataset_path, do_train, repeat_num=1, batch_size=32):
def create_dataset(dataset_path, do_train, repeat_num=1, infer_910=True, device_id=0, batch_size=32):
"""
create a train or evaluate cifar10 dataset for WideResnet
Args:
@ -31,13 +31,17 @@ def create_dataset(dataset_path, do_train, repeat_num=1, batch_size=32):
do_train(bool): whether dataset is used for train or eval.
repeat_num(int): the repeat times of dataset. Default: 1
batch_size(int): the batch size of dataset. Default: 32
infer_910(bool): infer 910 or infer 310. Default: True
device_id(int): infer 310 device_id. Default: 0
Returns:
dataset
"""
device_id = int(os.getenv('DEVICE_ID'))
device_num = int(os.getenv('RANK_SIZE'))
device_num = 1
device_id = device_id
if infer_910:
device_id = int(os.getenv('DEVICE_ID'))
device_num = int(os.getenv('RANK_SIZE'))
if do_train:
dataset_path = os.path.join(dataset_path, 'train')