Hello 分類サンプル¶
このサンプルでは、同期推論要求 API を使用して画像分類モデルを推論する方法を示します。サンプルを使用する前に、次の要件を参照してください。
入力と出力が 1 つだけのモデルがサポートされます。
このサンプルは、
core.read_modelでサポートされるすべてのファイル形式を受け入れます。サンプルをビルドするには、「サンプルを使ってみる」の「サンプル・アプリケーションをビルド」セクションにある手順を参照してください。
どのように動作するか¶
起動時に、サンプル・アプリケーションはコマンドライン・パラメーターを読み取り、入力データを準備し、指定されたモデルとイメージを OpenVINO™ ランタイムプラグインにロードし、同期推論を実行して、出力データを処理し、各ステップを標準出力ストリームに記録します。
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
# Copyright (C) 2018-2024 Intel Corporation
# SPDX-License-Identifier: Apache-2.0
import logging as log
import sys
import cv2
import numpy as np
import openvino as ov
def main():
log.basicConfig(format='[ %(levelname)s ] %(message)s', level=log.INFO, stream=sys.stdout)
# Parsing and validation of input arguments
if len(sys.argv) != 4:
log.info(f'Usage: {sys.argv[0]} <path_to_model> <path_to_image> <device_name>')
return 1
model_path = sys.argv[1]
image_path = sys.argv[2]
device_name = sys.argv[3]
# --------------------------- Step 1. Initialize OpenVINO Runtime Core ------------------------------------------------
log.info('Creating OpenVINO Runtime Core')
core = ov.Core()
# --------------------------- Step 2. Read a model --------------------------------------------------------------------
log.info(f'Reading the model: {model_path}')
# (.xml and .bin files) or (.onnx file)
model = core.read_model(model_path)
if len(model.inputs) != 1:
log.error('Sample supports only single input topologies')
return -1
if len(model.outputs) != 1:
log.error('Sample supports only single output topologies')
return -1
# --------------------------- Step 3. Set up input --------------------------------------------------------------------
# Read input image
image = cv2.imread(image_path)
# Add N dimension
input_tensor = np.expand_dims(image, 0)
# --------------------------- Step 4. Apply preprocessing -------------------------------------------------------------
ppp = ov.preprocess.PrePostProcessor(model)
_, h, w, _ = input_tensor.shape
# 1) Set input tensor information:
# - input() provides information about a single model input
# - reuse precision and shape from already available `input_tensor`
# - layout of data is 'NHWC'
ppp.input().tensor() \
.set_shape(input_tensor.shape) \
.set_element_type(ov.Type.u8) \
.set_layout(ov.Layout('NHWC')) # noqa: ECE001, N400
# 2) Adding explicit preprocessing steps:
# - apply linear resize from tensor spatial dims to model spatial dims
ppp.input().preprocess().resize(ov.preprocess.ResizeAlgorithm.RESIZE_LINEAR)
# 3) Here we suppose model has 'NCHW' layout for input
ppp.input().model().set_layout(ov.Layout('NCHW'))
# 4) Set output tensor information:
# - precision of tensor is supposed to be 'f32'
ppp.output().tensor().set_element_type(ov.Type.f32)
# 5) Apply preprocessing modifying the original 'model'
model = ppp.build()
# --------------------------- Step 5. Loading model to the device -----------------------------------------------------
log.info('Loading the model to the plugin')
compiled_model = core.compile_model(model, device_name)
# --------------------------- Step 6. Create infer request and do inference synchronously -----------------------------
log.info('Starting inference in synchronous mode')
results = compiled_model.infer_new_request({0: input_tensor})
# --------------------------- Step 7. Process output ------------------------------------------------------------------
predictions = next(iter(results.values()))
# Change a shape of a numpy.ndarray with results to get another one with one dimension
probs = predictions.reshape(-1)
# Get an array of 10 class IDs in descending order of probability
top_10 = np.argsort(probs)[-10:][::-1]
header = 'class_id probability'
log.info(f'Image path: {image_path}')
log.info('Top 10 results: ')
log.info(header)
log.info('-' * len(header))
for class_id in top_10:
probability_indent = ' ' * (len('class_id') - len(str(class_id)) + 1)
log.info(f'{class_id}{probability_indent}{probs[class_id]:.7f}')
log.info('')
# ----------------------------------------------------------------------------------------------------------------------
log.info('This sample is an API example, for any performance measurements please use the dedicated benchmark_app tool\n')
return 0
if __name__ == '__main__':
sys.exit(main())
// Copyright (C) 2018-2024 Intel Corporation
// SPDX-License-Identifier: Apache-2.0
//
#include <iterator>
#include <memory>
#include <sstream>
#include <string>
#include <vector>
// clang-format off
#include "openvino/openvino.hpp"
#include "samples/args_helper.hpp"
#include "samples/common.hpp"
#include "samples/classification_results.h"
#include "samples/slog.hpp"
#include "format_reader_ptr.h"
// clang-format on
/**
* @brief Main with support Unicode paths, wide strings
*/
int tmain(int argc, tchar* argv[]) {
try {
// -------- Get OpenVINO runtime version --------
slog::info << ov::get_openvino_version() << slog::endl;
// -------- Parsing and validation of input arguments --------
if (argc != 4) {
slog::info << "Usage : " << argv[0] << " <path_to_model> <path_to_image> <device_name>" << slog::endl;
return EXIT_FAILURE;
}
const std::string args = TSTRING2STRING(argv[0]);
const std::string model_path = TSTRING2STRING(argv[1]);
const std::string image_path = TSTRING2STRING(argv[2]);
const std::string device_name = TSTRING2STRING(argv[3]);
// -------- Step 1. Initialize OpenVINO Runtime Core --------
ov::Core core;
// -------- Step 2. Read a model --------
slog::info << "Loading model files: " << model_path << slog::endl;
std::shared_ptr<ov::Model> model = core.read_model(model_path);
printInputAndOutputsInfo(*model);
OPENVINO_ASSERT(model->inputs().size() == 1, "Sample supports models with 1 input only");
OPENVINO_ASSERT(model->outputs().size() == 1, "Sample supports models with 1 output only");
// -------- Step 3. Set up input
// Read input image to a tensor and set it to an infer request
// without resize and layout conversions
FormatReader::ReaderPtr reader(image_path.c_str());
if (reader.get() == nullptr) {
std::stringstream ss;
ss << "Image " + image_path + " cannot be read!";
throw std::logic_error(ss.str());
}
ov::element::Type input_type = ov::element::u8;
ov::Shape input_shape = {1, reader->height(), reader->width(), 3};
std::shared_ptr<unsigned char> input_data = reader->getData();
// just wrap image data by ov::Tensor without allocating of new memory
ov::Tensor input_tensor = ov::Tensor(input_type, input_shape, input_data.get());
const ov::Layout tensor_layout{"NHWC"};
// -------- Step 4. Configure preprocessing --------
ov::preprocess::PrePostProcessor ppp(model);
// 1) Set input tensor information:
// - input() provides information about a single model input
// - reuse precision and shape from already available `input_tensor`
// - layout of data is 'NHWC'
ppp.input().tensor().set_shape(input_shape).set_element_type(input_type).set_layout(tensor_layout);
// 2) Adding explicit preprocessing steps:
// - convert layout to 'NCHW' (from 'NHWC' specified above at tensor layout)
// - apply linear resize from tensor spatial dims to model spatial dims
ppp.input().preprocess().resize(ov::preprocess::ResizeAlgorithm::RESIZE_LINEAR);
// 4) Here we suppose model has 'NCHW' layout for input
ppp.input().model().set_layout("NCHW");
// 5) Set output tensor information:
// - precision of tensor is supposed to be 'f32'
ppp.output().tensor().set_element_type(ov::element::f32);
// 6) Apply preprocessing modifying the original 'model'
model = ppp.build();
// -------- Step 5. Loading a model to the device --------
ov::CompiledModel compiled_model = core.compile_model(model, device_name);
// -------- Step 6. Create an infer request --------
ov::InferRequest infer_request = compiled_model.create_infer_request();
// -----------------------------------------------------------------------------------------------------
// -------- Step 7. Prepare input --------
infer_request.set_input_tensor(input_tensor);
// -------- Step 8. Do inference synchronously --------
infer_request.infer();
// -------- Step 9. Process output
const ov::Tensor& output_tensor = infer_request.get_output_tensor();
// Print classification results
ClassificationResult classification_result(output_tensor, {image_path});
classification_result.show();
// -----------------------------------------------------------------------------------------------------
} catch (const std::exception& ex) {
std::cerr << ex.what() << std::endl;
return EXIT_FAILURE;
}
return EXIT_SUCCESS;
}
// Copyright (C) 2018-2024 Intel Corporation
// SPDX-License-Identifier: Apache-2.0
//
#include <opencv_c_wrapper.h>
#include <stdbool.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "openvino/c/openvino.h"
/**
* @brief Struct to store infer results
*/
struct infer_result {
size_t class_id;
float probability;
};
/**
* @brief Sort result by probability
* @param struct with infer results to sort
* @param result_size of the struct
* @return none
*/
int compare(const void* a, const void* b) {
const struct infer_result* sa = (const struct infer_result*)a;
const struct infer_result* sb = (const struct infer_result*)b;
if (sa->probability < sb->probability) {
return 1;
} else if ((sa->probability == sb->probability) && (sa->class_id > sb->class_id)) {
return 1;
} else if (sa->probability > sb->probability) {
return -1;
}
return 0;
}
void infer_result_sort(struct infer_result* results, size_t result_size) {
qsort(results, result_size, sizeof(struct infer_result), compare);
}
/**
* @brief Convert output tensor to infer result struct for processing results
* @param tensor of output tensor
* @param result_size of the infer result
* @return struct infer_result
*/
struct infer_result* tensor_to_infer_result(ov_tensor_t* tensor, size_t* result_size) {
ov_shape_t output_shape = {0};
ov_status_e status = ov_tensor_get_shape(tensor, &output_shape);
if (status != OK)
return NULL;
*result_size = output_shape.dims[1];
struct infer_result* results = (struct infer_result*)malloc(sizeof(struct infer_result) * (*result_size));
if (!results)
return NULL;
void* data = NULL;
status = ov_tensor_data(tensor, &data);
if (status != OK) {
free(results);
return NULL;
}
float* float_data = (float*)(data);
size_t i;
for (i = 0; i < *result_size; ++i) {
results[i].class_id = i;
results[i].probability = float_data[i];
}
ov_shape_free(&output_shape);
return results;
}
/**
* @brief Print results of infer
* @param results of the infer results
* @param result_size of the struct of classification results
* @param img_path image path
* @return none
*/
void print_infer_result(struct infer_result* results, size_t result_size, const char* img_path) {
printf("\nImage %s\n", img_path);
printf("\nclassid probability\n");
printf("------- -----------\n");
size_t i;
for (i = 0; i < result_size; ++i) {
printf("%zu %f\n", results[i].class_id, results[i].probability);
}
}
void print_model_input_output_info(ov_model_t* model) {
char* friendly_name = NULL;
ov_model_get_friendly_name(model, &friendly_name);
printf("[INFO] model name: %s \n", friendly_name);
ov_free(friendly_name);
}
#define CHECK_STATUS(return_status) \
if (return_status != OK) { \
fprintf(stderr, "[ERROR] return status %d, line %d\n", return_status, __LINE__); \
goto err; \
}
int main(int argc, char** argv) {
// -------- Check input parameters --------
if (argc != 4) {
printf("Usage : ./hello_classification_c <path_to_model> <path_to_image> "
"<device_name>\n");
return EXIT_FAILURE;
}
ov_core_t* core = NULL;
ov_model_t* model = NULL;
ov_tensor_t* tensor = NULL;
ov_preprocess_prepostprocessor_t* preprocess = NULL;
ov_preprocess_input_info_t* input_info = NULL;
ov_model_t* new_model = NULL;
ov_preprocess_input_tensor_info_t* input_tensor_info = NULL;
ov_preprocess_preprocess_steps_t* input_process = NULL;
ov_preprocess_input_model_info_t* p_input_model = NULL;
ov_preprocess_output_info_t* output_info = NULL;
ov_preprocess_output_tensor_info_t* output_tensor_info = NULL;
ov_compiled_model_t* compiled_model = NULL;
ov_infer_request_t* infer_request = NULL;
ov_tensor_t* output_tensor = NULL;
struct infer_result* results = NULL;
ov_layout_t* input_layout = NULL;
ov_layout_t* model_layout = NULL;
ov_shape_t input_shape;
ov_output_const_port_t* output_port = NULL;
ov_output_const_port_t* input_port = NULL;
// -------- Get OpenVINO runtime version --------
ov_version_t version;
CHECK_STATUS(ov_get_openvino_version(&version));
printf("---- OpenVINO INFO----\n");
printf("Description : %s \n", version.description);
printf("Build number: %s \n", version.buildNumber);
ov_version_free(&version);
// -------- Parsing and validation of input arguments --------
const char* input_model = argv[1];
const char* input_image_path = argv[2];
const char* device_name = argv[3];
// -------- Step 1. Initialize OpenVINO Runtime Core --------
CHECK_STATUS(ov_core_create(&core));
// -------- Step 2. Read a model --------
printf("[INFO] Loading model files: %s\n", input_model);
CHECK_STATUS(ov_core_read_model(core, input_model, NULL, &model));
print_model_input_output_info(model);
CHECK_STATUS(ov_model_const_output(model, &output_port));
if (!output_port) {
fprintf(stderr, "[ERROR] Sample supports models with 1 output only %d\n", __LINE__);
goto err;
}
CHECK_STATUS(ov_model_const_input(model, &input_port));
if (!input_port) {
fprintf(stderr, "[ERROR] Sample supports models with 1 input only %d\n", __LINE__);
goto err;
}
// -------- Step 3. Set up input
c_mat_t img;
image_read(input_image_path, &img);
ov_element_type_e input_type = U8;
int64_t dims[4] = {1, (size_t)img.mat_height, (size_t)img.mat_width, 3};
ov_shape_create(4, dims, &input_shape);
CHECK_STATUS(ov_tensor_create_from_host_ptr(input_type, input_shape, img.mat_data, &tensor));
// -------- Step 4. Configure preprocessing --------
CHECK_STATUS(ov_preprocess_prepostprocessor_create(model, &preprocess));
CHECK_STATUS(ov_preprocess_prepostprocessor_get_input_info_by_index(preprocess, 0, &input_info));
CHECK_STATUS(ov_preprocess_input_info_get_tensor_info(input_info, &input_tensor_info));
CHECK_STATUS(ov_preprocess_input_tensor_info_set_from(input_tensor_info, tensor));
const char* input_layout_desc = "NHWC";
CHECK_STATUS(ov_layout_create(input_layout_desc, &input_layout));
CHECK_STATUS(ov_preprocess_input_tensor_info_set_layout(input_tensor_info, input_layout));
CHECK_STATUS(ov_preprocess_input_info_get_preprocess_steps(input_info, &input_process));
CHECK_STATUS(ov_preprocess_preprocess_steps_resize(input_process, RESIZE_LINEAR));
CHECK_STATUS(ov_preprocess_input_info_get_model_info(input_info, &p_input_model));
const char* model_layout_desc = "NCHW";
CHECK_STATUS(ov_layout_create(model_layout_desc, &model_layout));
CHECK_STATUS(ov_preprocess_input_model_info_set_layout(p_input_model, model_layout));
CHECK_STATUS(ov_preprocess_prepostprocessor_get_output_info_by_index(preprocess, 0, &output_info));
CHECK_STATUS(ov_preprocess_output_info_get_tensor_info(output_info, &output_tensor_info));
CHECK_STATUS(ov_preprocess_output_set_element_type(output_tensor_info, F32));
CHECK_STATUS(ov_preprocess_prepostprocessor_build(preprocess, &new_model));
// -------- Step 5. Loading a model to the device --------
CHECK_STATUS(ov_core_compile_model(core, new_model, device_name, 0, &compiled_model));
// -------- Step 6. Create an infer request --------
CHECK_STATUS(ov_compiled_model_create_infer_request(compiled_model, &infer_request));
// -------- Step 7. Prepare input --------
CHECK_STATUS(ov_infer_request_set_input_tensor_by_index(infer_request, 0, tensor));
// -------- Step 8. Do inference synchronously --------
CHECK_STATUS(ov_infer_request_infer(infer_request));
// -------- Step 9. Process output
CHECK_STATUS(ov_infer_request_get_output_tensor_by_index(infer_request, 0, &output_tensor));
// Print classification results
size_t results_num;
results = tensor_to_infer_result(output_tensor, &results_num);
infer_result_sort(results, results_num);
size_t top = 10;
if (top > results_num) {
top = results_num;
}
printf("\nTop %zu results:\n", top);
print_infer_result(results, top, input_image_path);
// -------- free allocated resources --------
err:
free(results);
image_free(&img);
ov_shape_free(&input_shape);
ov_output_const_port_free(output_port);
ov_output_const_port_free(input_port);
if (output_tensor)
ov_tensor_free(output_tensor);
if (infer_request)
ov_infer_request_free(infer_request);
if (compiled_model)
ov_compiled_model_free(compiled_model);
if (input_layout)
ov_layout_free(input_layout);
if (model_layout)
ov_layout_free(model_layout);
if (output_tensor_info)
ov_preprocess_output_tensor_info_free(output_tensor_info);
if (output_info)
ov_preprocess_output_info_free(output_info);
if (p_input_model)
ov_preprocess_input_model_info_free(p_input_model);
if (input_process)
ov_preprocess_preprocess_steps_free(input_process);
if (input_tensor_info)
ov_preprocess_input_tensor_info_free(input_tensor_info);
if (input_info)
ov_preprocess_input_info_free(input_info);
if (preprocess)
ov_preprocess_prepostprocessor_free(preprocess);
if (new_model)
ov_model_free(new_model);
if (tensor)
ov_tensor_free(tensor);
if (model)
ov_model_free(model);
if (core)
ov_core_free(core);
return EXIT_SUCCESS;
}
各サンプルの明示的な説明は、「OpenVINO™ をアプリケーションと統合」の統合ステップを確認してください。
実行¶
python hello_classification.py <path_to_model> <path_to_image> <device_name>
hello_classification <path_to_model> <path_to_image> <device_name>
hello_classification_c <path_to_model> <path_to_image> <device_name>
サンプルを実行するには、モデルとイメージを指定する必要があります。
TensorFlow Zoo、Hugging Face、TensorFlow Hub などのモデル・リポジトリーから推論タスクに固有のモデルを取得できます。
ストレージで利用可能なメディア・ファイル・コレクションの画像を使用できます。
注
OpenVINO™ ツールキットのサンプルとデモは、デフォルトでは BGR チャネル順序での入力を想定しています。RGB 順序で動作するようにモデルをトレーニングした場合は、サンプルまたはデモ・アプリケーションでデフォルトのチャネル順序を手動で再配置するか、
reverse_input_channels引数を指定したモデル変換 API を使用してモデルを再変換する必要があります。引数の詳細については、前処理計算の埋め込みの入力チャネルを反転するときセクションを参照してください。トレーニングされたモデルでサンプルを実行する前に、モデル変換 API を使用してモデルが中間表現 (IR) 形式 (*.xml + *.bin) に変換されていることを確認してください。
このサンプルは、前処理を必要としない ONNX 形式 (.onnx) のモデルを受け入れます。
例¶
事前トレーニングされたモデルをダウンロードします。
-
以下を使用して変換できます。
import openvino as ov ov_model = ov.convert_model('./models/alexnet') # or, when model is a Python model object ov_model = ov.convert_model(alexnet)
ovc ./models/alexnet -
GPU上のモデルを使用して、画像の推論を実行します。
例:python hello_classification.py ./models/alexnet/alexnet.xml ./images/banana.jpg GPUhello_classification ./models/googlenet-v1.xml ./images/car.bmp GPUhello_classification_c alexnet.xml ./opt/intel/openvino/samples/scripts/car.png GPU
サンプルの出力¶
サンプル・アプリケーションは、各ステップを標準出力ストリームに記録し、上位 10 の推論結果を出力します。
[ INFO ] Creating OpenVINO Runtime Core
[ INFO ] Reading the model: /models/alexnet/alexnet.xml
[ INFO ] Loading the model to the plugin
[ INFO ] Starting inference in synchronous mode
[ INFO ] Image path: /images/banana.jpg
[ INFO ] Top 10 results:
[ INFO ] class_id probability
[ INFO ] --------------------
[ INFO ] 954 0.9703885
[ INFO ] 666 0.0219518
[ INFO ] 659 0.0033120
[ INFO ] 435 0.0008246
[ INFO ] 809 0.0004433
[ INFO ] 502 0.0003852
[ INFO ] 618 0.0002906
[ INFO ] 910 0.0002848
[ INFO ] 951 0.0002427
[ INFO ] 961 0.0002213
[ INFO ]
[ INFO ] This sample is an API example, for any performance measurements please use the dedicated benchmark_app tool
アプリケーションは上位 10 の推論結果を出力します。
[ INFO ] OpenVINO Runtime version ......... <version>
[ INFO ] Build ........... <build>
[ INFO ]
[ INFO ] Loading model files: /models/googlenet-v1.xml
[ INFO ] model name: GoogleNet
[ INFO ] inputs
[ INFO ] input name: data
[ INFO ] input type: f32
[ INFO ] input shape: {1, 3, 224, 224}
[ INFO ] outputs
[ INFO ] output name: prob
[ INFO ] output type: f32
[ INFO ] output shape: {1, 1000}
Top 10 results:
Image /images/car.bmp
classid probability
------- -----------
656 0.8139648
654 0.0550537
468 0.0178375
436 0.0165405
705 0.0111694
817 0.0105820
581 0.0086823
575 0.0077515
734 0.0064468
785 0.0043983
アプリケーションは上位 10 の推論結果を出力します。
Top 10 results:
Image /opt/intel/openvino/samples/scripts/car.png
classid probability
------- -----------
656 0.666479
654 0.112940
581 0.068487
874 0.033385
436 0.026132
817 0.016731
675 0.010980
511 0.010592
569 0.008178
717 0.006336
This sample is an API example, for any performance measurements use the dedicated benchmark_app tool.