openvino/inference-engine/samples/validation_app/image_decoder.cpp

142 lines
4.9 KiB
C++

/*
// Copyright (c) 2018 Intel Corporation
//
// 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 <utility>
#include "image_decoder.hpp"
#include "details/ie_exception.hpp"
#include <vector>
#include <string>
#include <map>
#include <iostream>
#include <opencv2/core/core.hpp>
#include <opencv2/highgui/highgui.hpp>
#include <opencv2/imgproc/imgproc.hpp>
using namespace cv;
int getLoadModeForChannels(int channels, int base) {
switch (channels) {
case 1:
return base | CV_LOAD_IMAGE_GRAYSCALE;
case 3:
return base | CV_LOAD_IMAGE_COLOR;
}
return base | CV_LOAD_IMAGE_UNCHANGED;
}
template <class T>
cv::Size addToBlob(std::string name, int batch_pos, Blob& blob, PreprocessingOptions preprocessingOptions) {
SizeVector blobSize = blob.dims();
int width = static_cast<int>(blobSize[0]);
int height = static_cast<int>(blobSize[1]);
int channels = static_cast<int>(blobSize[2]);
T* blob_data = static_cast<T*>(blob.buffer());
Mat orig_image, result_image;
int loadMode = getLoadModeForChannels(channels, 0);
std::string tryName = name;
// TODO This is a dirty hack to support VOC2007 (where no file extension is put into annotation).
// Rewrite.
if (name.find('.') == -1) tryName = name + ".JPEG";
orig_image = imread(tryName, loadMode);
if (orig_image.empty()) {
THROW_IE_EXCEPTION << "Cannot open image file: " << tryName;
}
// Preprocessing the image
Size res = orig_image.size();
if (preprocessingOptions.resizeCropPolicy == ResizeCropPolicy::Resize) {
cv::resize(orig_image, result_image, Size(width, height));
} else if (preprocessingOptions.resizeCropPolicy == ResizeCropPolicy::ResizeThenCrop) {
Mat resized_image;
cv::resize(orig_image, resized_image, Size(preprocessingOptions.resizeBeforeCropX, preprocessingOptions.resizeBeforeCropY));
size_t cx = preprocessingOptions.resizeBeforeCropX / 2;
size_t cy = preprocessingOptions.resizeBeforeCropY / 2;
cv::Rect cropRect(cx - width / 2, cy - height / 2, width, height);
result_image = resized_image(cropRect);
} else if (preprocessingOptions.resizeCropPolicy == ResizeCropPolicy::DoNothing) {
// No image preprocessing to be done here
result_image = orig_image;
} else {
THROW_IE_EXCEPTION << "Unsupported ResizeCropPolicy value";
}
float scaleFactor = preprocessingOptions.scaleValuesTo01 ? 255.0 : 1.0;
for (int c = 0; c < channels; c++) {
for (int h = 0; h < height; h++) {
for (int w = 0; w < width; w++) {
blob_data[batch_pos * channels * width * height + c * width * height + h * width + w] =
static_cast<T>(result_image.at<cv::Vec3b>(h, w)[c] / scaleFactor);
}
}
}
return res;
}
std::map<std::string, cv::Size> convertToBlob(std::vector<std::string> names, int batch_pos, Blob& blob, PreprocessingOptions preprocessingOptions) {
if (blob.buffer() == nullptr) {
THROW_IE_EXCEPTION << "Blob was not allocated";
}
std::function<cv::Size(std::string, int, Blob&, PreprocessingOptions)> add_func;
switch (blob.precision()) {
case Precision::FP32:
add_func = &addToBlob<float>;
break;
case Precision::FP16:
case Precision::Q78:
case Precision::I16:
case Precision::U16:
add_func = &addToBlob<short>;
break;
default:
add_func = &addToBlob<uint8_t>;
}
std::map<std::string, Size> res;
for (int b = 0; b < names.size(); b++) {
std::string name = names[b];
Size orig_size = add_func(name, batch_pos + b, blob, preprocessingOptions);
res.insert(std::pair<std::string, Size>(name, orig_size));
}
return res;
}
Size ImageDecoder::loadToBlob(std::string name, Blob& blob, PreprocessingOptions preprocessingOptions) {
std::vector<std::string> names = { name };
return loadToBlob(names, blob, preprocessingOptions).at(name);
}
std::map<std::string, cv::Size> ImageDecoder::loadToBlob(std::vector<std::string> names, Blob& blob, PreprocessingOptions preprocessingOptions) {
return convertToBlob(names, 0, blob, preprocessingOptions);
}
Size ImageDecoder::insertIntoBlob(std::string name, int batch_pos, Blob& blob, PreprocessingOptions preprocessingOptions) {
return convertToBlob({ name }, batch_pos, blob, preprocessingOptions).at(name);
}