| /** |
| * Copyright (C) 2020 Savoir-faire Linux Inc. |
| * |
| * Author: Aline Gondim Santos <aline.gondimsantos@savoirfairelinux.com> |
| * |
| * This program is free software; you can redistribute it and/or modify |
| * it under the terms of the GNU General Public License as published by |
| * the Free Software Foundation; either version 3 of the License, or |
| * (at your option) any later version. |
| * |
| * This program is distributed in the hope that it will be useful, |
| * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| * GNU General Public License for more details. |
| * |
| * You should have received a copy of the GNU General Public License |
| * along with this program; if not, write to the Free Software |
| * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 |
| * USA. |
| */ |
| |
| #include "pluginProcessor.h" |
| // System includes |
| #include <algorithm> |
| #include <cstring> |
| // OpenCV headers |
| #include <opencv2/core.hpp> |
| #include <opencv2/imgcodecs.hpp> |
| #include <opencv2/imgproc.hpp> |
| // Logger |
| #include <pluglog.h> |
| |
| extern "C" { |
| #include <libavutil/display.h> |
| } |
| const char sep = separator(); |
| |
| const std::string TAG = "FORESEG"; |
| |
| namespace jami { |
| |
| PluginProcessor::PluginProcessor(const std::string& dataPath, const std::string& model, const std::string& backgroundImage, bool acc) |
| { |
| activateAcc_ = acc; |
| initModel(dataPath+sep+"model/"+model); |
| setBackgroundImage(backgroundImage); |
| } |
| |
| PluginProcessor::~PluginProcessor() |
| { |
| Plog::log(Plog::LogPriority::INFO, TAG, "~pluginprocessor"); |
| if (session_) |
| delete session_; |
| } |
| |
| void |
| PluginProcessor::setBackgroundImage(const std::string& backgroundPath) |
| { |
| cv::Size size = cv::Size {0, 0}; |
| |
| if (!backgroundImage.empty()) |
| size = backgroundImage.size(); |
| |
| cv::Mat newBackgroundImage = cv::imread(backgroundPath); |
| if (newBackgroundImage.cols == 0) { |
| Plog::log(Plog::LogPriority::ERR, TAG, "Background image not Loaded"); |
| } else { |
| Plog::log(Plog::LogPriority::INFO, TAG, "Background image Loaded"); |
| cv::cvtColor(newBackgroundImage, newBackgroundImage, cv::COLOR_BGR2RGB); |
| newBackgroundImage.convertTo(newBackgroundImage, CV_32FC3); |
| if (size.height) { |
| cv::resize(newBackgroundImage, newBackgroundImage, size); |
| backgroundRotation = 0; |
| } |
| backgroundImage = newBackgroundImage.clone(); |
| newBackgroundImage.release(); |
| hasBackground_ = true; |
| } |
| } |
| |
| void |
| PluginProcessor::initModel(const std::string& modelPath) |
| { |
| try { |
| auto allocator_info = Ort::MemoryInfo::CreateCpu(OrtArenaAllocator, OrtMemTypeDefault); |
| input_tensor_ = Ort::Value::CreateTensor<float>(allocator_info, input_image_.data(), input_image_.size(), input_shape_.data(), input_shape_.size()); |
| output_tensor_ = Ort::Value::CreateTensor<float>(allocator_info, results_.data(), results_.size(), output_shape_.data(), output_shape_.size()); |
| sessOpt_ = Ort::SessionOptions(); |
| |
| #ifdef NVIDIA |
| if (activateAcc_) |
| Ort::ThrowOnError(OrtSessionOptionsAppendExecutionProvider_CUDA(sessOpt_, 0)); |
| #endif |
| #ifdef ANDROID |
| if (activateAcc_) |
| Ort::ThrowOnError(OrtSessionOptionsAppendExecutionProvider_Nnapi(sessOpt_, 0)); |
| #endif |
| |
| sessOpt_.SetGraphOptimizationLevel(GraphOptimizationLevel::ORT_ENABLE_ALL); |
| #ifdef WIN32 |
| std::wstring wsTmp(modelPath.begin(), modelPath.end()); |
| session_ = new Ort::Session(env, wsTmp.c_str(), sessOpt_); |
| #else |
| session_ = new Ort::Session(env, modelPath.c_str(), sessOpt_); |
| #endif |
| isAllocated_ = true; |
| } catch (std::exception& e) { |
| Plog::log(Plog::LogPriority::ERR, TAG, e.what()); |
| } |
| std::ostringstream oss; |
| oss << "Model is allocated " << isAllocated_; |
| Plog::log(Plog::LogPriority::INFO, TAG, oss.str()); |
| } |
| |
| bool |
| PluginProcessor::isAllocated() |
| { |
| return isAllocated_; |
| } |
| |
| void |
| PluginProcessor::feedInput(const cv::Mat& frame) |
| { |
| cv::Mat temp(frame.rows, frame.cols, CV_32FC3, input_image_.data()); |
| frame.convertTo(temp, CV_32FC3); |
| } |
| |
| int |
| PluginProcessor::getBackgroundRotation() |
| { |
| return backgroundRotation; |
| } |
| |
| void |
| PluginProcessor::setBackgroundRotation(int angle) |
| { |
| if (backgroundRotation != angle && (backgroundRotation - angle) != 0) { |
| rotateFrame(backgroundRotation - angle, backgroundImage); |
| backgroundRotation = angle; |
| } |
| } |
| |
| void |
| PluginProcessor::computePredictions() |
| { |
| if (count == 0) { |
| // Run the graph |
| session_->Run(Ort::RunOptions{nullptr}, input_names, &input_tensor_, 1, output_names, &output_tensor_, 1); |
| computedMask = std::vector(results_.begin(), results_.end()); |
| } |
| } |
| |
| void |
| PluginProcessor::printMask() |
| { |
| for (size_t i = 0; i < computedMask.size(); i++) { |
| // Log the predictions |
| std::ostringstream oss; |
| oss << "\nclass: " << computedMask[i] << std::endl; |
| Plog::log(Plog::LogPriority::INFO, TAG, oss.str()); |
| } |
| } |
| |
| void |
| PluginProcessor::resetInitValues(const cv::Size& modelInputSize) |
| { |
| previousMasks[0] = cv::Mat(modelInputSize.height, modelInputSize.width, CV_32FC1, double(0.)); |
| previousMasks[1] = cv::Mat(modelInputSize.height, modelInputSize.width, CV_32FC1, double(0.)); |
| kSize = cv::Size(modelInputSize.width * kernelSize, modelInputSize.height * kernelSize); |
| if (kSize.height % 2 == 0) { |
| kSize.height -= 1; |
| } |
| if (kSize.width % 2 == 0) { |
| kSize.width -= 1; |
| } |
| count = 0; |
| grabCutMode = cv::GC_INIT_WITH_MASK; |
| grabCutIterations = 5; |
| } |
| |
| void |
| copyByLine(uchar* frameData, uchar* applyMaskData, const int lineSize, cv::Size size) |
| { |
| if (3 * size.width == lineSize) { |
| std::memcpy(frameData, applyMaskData, size.height * size.width * 3); |
| } else { |
| int rows = size.height; |
| int offset = 0; |
| int maskoffset = 0; |
| for (int i = 0; i < rows; i++) { |
| std::memcpy(frameData + offset, applyMaskData + maskoffset, lineSize); |
| offset += lineSize; |
| maskoffset += 3 * size.width; |
| } |
| } |
| } |
| |
| void |
| PluginProcessor::drawMaskOnFrame( |
| cv::Mat& frame, cv::Mat& frameReduced, std::vector<float> computedMask, int lineSize, int angle) |
| { |
| if (computedMask.empty()) { |
| return; |
| } |
| |
| if (count == 0) { |
| int maskSize = static_cast<int>(std::sqrt(computedMask.size())); |
| cv::Mat maskImg(maskSize, maskSize, CV_32FC1, computedMask.data()); |
| cv::Mat* applyMask = &frameReduced; |
| |
| rotateFrame(-angle, maskImg); |
| cv::resize(maskImg, maskImg, cv::Size(frameReduced.cols, frameReduced.rows)); |
| |
| double m, M; |
| cv::minMaxLoc(maskImg, &m, &M); |
| |
| if (M < 2) { // avoid detection if there is any one in frame |
| maskImg = 0. * maskImg; |
| } else { |
| for (int i = 0; i < maskImg.cols; i++) { |
| for (int j = 0; j < maskImg.rows; j++) { |
| maskImg.at<float>(j, i) = (maskImg.at<float>(j, i) - m) / (M - m); |
| |
| if (maskImg.at<float>(j, i) < 0.4) |
| maskImg.at<float>(j, i) = 0.; |
| else if (maskImg.at<float>(j, i) < 0.7) { |
| float value = maskImg.at<float>(j, i) * smoothFactors[0] |
| + previousMasks[0].at<float>(j, i) * smoothFactors[1] |
| + previousMasks[1].at<float>(j, i) * smoothFactors[2]; |
| maskImg.at<float>(j, i) = 0.; |
| if (value > 0.7) |
| maskImg.at<float>(j, i) = 1.; |
| } else |
| maskImg.at<float>(j, i) = 1.; |
| } |
| } |
| } |
| if (cv::countNonZero(maskImg) != 0) { |
| cv::Mat dilate; |
| cv::dilate(maskImg, |
| dilate, |
| cv::getStructuringElement(cv::MORPH_ELLIPSE, kSize), |
| cv::Point(-1, -1), |
| 2); |
| cv::erode(maskImg, |
| maskImg, |
| cv::getStructuringElement(cv::MORPH_ELLIPSE, kSize), |
| cv::Point(-1, -1), |
| 2); |
| for (int i = 0; i < maskImg.cols; i++) { |
| for (int j = 0; j < maskImg.rows; j++) { |
| if (dilate.at<float>(j, i) != maskImg.at<float>(j, i)) |
| maskImg.at<float>(j, i) = grabcutClass; |
| } |
| } |
| maskImg.convertTo(maskImg, CV_8UC1); |
| applyMask->convertTo(*applyMask, CV_8UC1); |
| cv::Rect rect(1, 1, maskImg.rows, maskImg.cols); |
| cv::grabCut(*applyMask, |
| maskImg, |
| rect, |
| bgdModel, |
| fgdModel, |
| grabCutIterations, |
| grabCutMode); |
| |
| grabCutMode = cv::GC_EVAL; |
| grabCutIterations = 1; |
| |
| maskImg = maskImg & 1; |
| maskImg.convertTo(maskImg, CV_32FC1); |
| maskImg *= 255.; |
| GaussianBlur(maskImg, maskImg, cv::Size(7, 7), 0); // float mask from 0 to 255. |
| maskImg = maskImg / 255.; |
| } |
| previousMasks[1] = previousMasks[0].clone(); |
| previousMasks[0] = maskImg.clone(); |
| } |
| |
| cv::Mat roiMaskImg = previousMasks[0].clone(); |
| cv::Mat roiMaskImgComplementary = 1. - roiMaskImg; // mask from 1. to 0 |
| |
| std::vector<cv::Mat> channels; |
| std::vector<cv::Mat> channelsComplementary; |
| |
| channels.emplace_back(roiMaskImg); |
| channels.emplace_back(roiMaskImg); |
| channels.emplace_back(roiMaskImg); |
| channelsComplementary.emplace_back(roiMaskImgComplementary); |
| channelsComplementary.emplace_back(roiMaskImgComplementary); |
| channelsComplementary.emplace_back(roiMaskImgComplementary); |
| |
| cv::merge(channels, roiMaskImg); |
| cv::merge(channelsComplementary, roiMaskImgComplementary); |
| |
| cv::Mat output; |
| frameReduced.convertTo(output, roiMaskImg.type()); |
| output = output.mul(roiMaskImg); |
| output += backgroundImage.mul(roiMaskImgComplementary); |
| output.convertTo(output, frameReduced.type()); |
| |
| cv::resize(output, output, cv::Size(frame.cols, frame.rows)); |
| |
| copyByLine(frame.data, output.data, lineSize, cv::Size(frame.cols, frame.rows)); |
| count++; |
| count = count % frameCount; |
| } |
| |
| void |
| PluginProcessor::rotateFrame(int angle, cv::Mat& mat) |
| { |
| if (angle == -90) |
| cv::rotate(mat, mat, cv::ROTATE_90_COUNTERCLOCKWISE); |
| else if (std::abs(angle) == 180) |
| cv::rotate(mat, mat, cv::ROTATE_180); |
| else if (angle == 90) |
| cv::rotate(mat, mat, cv::ROTATE_90_CLOCKWISE); |
| } |
| |
| bool |
| PluginProcessor::hasBackground() const |
| { |
| return hasBackground_; |
| } |
| } // namespace jami |