Fixed rescaling issue & score for COCO val.

7ae8d776 · Gines Hidalgo Martinez · bc55f86c · 7ae8d776 · 7ae8d776 · 7ae8d776
46 changed file
--- a/3rdparty/caffe/install_caffe_if_cuda8.sh
+++ b/3rdparty/caffe/install_caffe_if_cuda8.sh
@@ -57,10 +57,11 @@ sudo apt-get --assume-yes install --no-install-recommends libboost-all-dev
    sudo apt-get --assume-yes install libgflags-dev libgoogle-glog-dev liblmdb-dev
 # fi
 # Python libs
-sudo apt-get --assume-yes install python-setuptools python-dev build-essential
-sudo easy_install pip
-sudo -H pip install --upgrade numpy protobuf
+sudo apt-get --assume-yes install python-setuptools python-pip python-dev build-essential
 sudo apt-get install python-numpy
+sudo easy_install pip
+sudo -H pip install numpy protobuf
+# sudo -H pip install --upgrade numpy protobuf
 sudo ldconfig
 # OpenCV 2.4
 # sudo apt-get --assume-yes install libopencv-dev

--- a/doc/installation.md
+++ b/doc/installation.md
@@ -57,7 +57,7 @@ Recommended installation method. It is simpler and it offers many more customiza
 ### Installation - Script Compilation
 **Highly important**: This script only works with CUDA 8 and Ubuntu 14 or 16. Otherwise, see [doc/installation_cmake.md](installation_cmake.md) or [Installation - Manual Compilation](#installation---manual-compilation).
 1. Required: CUDA, cuDNN, OpenCV and Atlas must be already installed on your machine.
-    1. [CUDA](https://developer.nvidia.com/cuda-downloads) must be installed. You should reboot your machine after installing CUDA.
+    1. [CUDA](https://developer.nvidia.com/cuda-80-ga2-download-archive) must be installed. You should reboot your machine after installing CUDA.
    2. [cuDNN](https://developer.nvidia.com/cudnn): Once you have downloaded it, just unzip it and copy (merge) the contents on the CUDA folder, e.g. `/usr/local/cuda-8.0/`. Note: We found OpenPose working ~10% faster with cuDNN 5.1 compared to cuDNN 6. Otherwise, check [Compiling without cuDNN](#compiling-without-cudnn).
    3. OpenCV can be installed with `apt-get install libopencv-dev`. If you have compiled OpenCV 3 by your own, follow [Manual Compilation](#manual-compilation). After both Makefile.config files have been generated, edit them and uncomment the line `# OPENCV_VERSION := 3`. You might alternatively modify all `Makefile.config.UbuntuXX` files and then run the scripts in step 2.
    4. In addition, OpenCV 3 does not incorporate the `opencv_contrib` module by default. Assuming you have OpenCV 3 compiled with the contrib module and you want to use it, append `opencv_contrib` at the end of the line `LIBRARIES += opencv_core opencv_highgui opencv_imgproc` in the `Makefile` file.
@@ -134,7 +134,7 @@ You just need to remove the OpenPose folder, by default called `openpose/`. E.g.
 ### Installation - Library
 1. Install the pre-requisites:
    1. Microsoft Visual Studio (VS) 2015 Enterprise Update 3. If Visual Studio 2017 Community is desired, we do not support it, but it might be compiled by firstly [enabling CUDA 8.0 in VS2017](https://stackoverflow.com/questions/43745099/using-cuda-with-visual-studio-2017?answertab=active#tab-top). VS Enterprise Update 1 will give some compiler errors and VS 2015 Community has not been tested.
-    2. [CUDA 8](https://developer.nvidia.com/cuda-downloads): Install it on the default location, `C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0`. Otherwise, modify the Visual Studio project solution accordingly. Install CUDA 8.0 after Visual Studio 2015 is installed to assure that the CUDA installation will generate all necessary files for VS. If CUDA was already installed, re-install it after installing VS!
+    2. [CUDA 8](https://developer.nvidia.com/cuda-80-ga2-download-archive): Install it on the default location, `C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0`. Otherwise, modify the Visual Studio project solution accordingly. Install CUDA 8.0 after Visual Studio 2015 is installed to assure that the CUDA installation will generate all necessary files for VS. If CUDA was already installed, re-install it after installing VS!
    3. [cuDNN 5.1](https://developer.nvidia.com/cudnn): Once you have downloaded it, just unzip it and copy (merge) the contents on the CUDA folder, `C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0`.
 2. Download the OpenPose dependencies and models (body, face and hand models) by double-clicking on `{openpose_path}\windows\download_3rdparty_and_models.bat`. Alternatively, you might prefer to download them manually:
    - Models:

--- a/doc/release_notes.md
+++ b/doc/release_notes.md
@@ -113,7 +113,7 @@ OpenPose Library - Release Notes



-## Current version (future OpenPose 2.0.0alpha)
+## Current version (future OpenPose 1.2.0alpha)
 1. Main improvements:
    1. Added IP camera support.
    2. Output images can have the input size, OpenPose able to change its size for each image and not required fixed size anymore.
@@ -128,7 +128,7 @@ OpenPose Library - Release Notes
    8. Added example for OpenPose for user asynchronous output and cleaned all `tutorial_wrapper/` examples.
    9. Added `-1` option for `net_resolution` in order to auto-select the best possible aspect ratio given the user input.
    10. Net resolution can be dynamically changed (e.g. for images with different size).
-    11. Added example to add functionality to OpenPose.
+    11. Added example to add functionality/modules to OpenPose.
 2. Functions or parameters renamed:
    1. OpenPose able to change its size and initial size dynamically:
        1. Flag `resolution` renamed as `output_resolution`.
@@ -140,5 +140,6 @@ OpenPose Library - Release Notes
        1. Changed several functions on `core/`, `pose/`, `face/`, and `hand/` modules.
    3. `CPU_ONLY` changed by `USE_CUDA` to keep format.
 3. Main bugs fixed:
-    1. Ubuntu installer script now works even if Python pip was not installed previously.
-    2. Flags to set first and last frame as well as jumping frames backward and forward now works on image directory reader.
+    1. Scaling resize issue fixed: ~1-pixel offset due to not considering 0-based indexes.
+    2. Ubuntu installer script now works even if Python pip was not installed previously.
+    3. Flags to set first and last frame as well as jumping frames backward and forward now works on image directory reader.
--- a/examples/tests/pose_accuracy_coco_test.sh
+++ b/examples/tests/pose_accuracy_coco_test.sh
@@ -22,13 +22,15 @@ OP_BIN=./build/examples/openpose/openpose.bin

    # 1 scale
 $OP_BIN --image_dir $IMAGE_FOLDER --write_coco_json ${JSON_FOLDER}1.json --no_display --render_pose 0 --frame_last 3558
-# --output_resolution "1280x720"
+
+	# 1 scale - Debugging
+# $OP_BIN --image_dir $IMAGE_FOLDER --write_coco_json ${JSON_FOLDER}1.json --no_display --frame_last 3558 --write_images ~/Desktop/CppValidation/

 #     # 3 scales
-# $OP_BIN --image_dir $IMAGE_FOLDER --write_coco_json ${JSON_FOLDER}1_3.json --no_display --render_pose 0 --scale_number 3 --scale_gap 0.25 --frame_last 3558 --output_resolution "1280x720"
+# $OP_BIN --image_dir $IMAGE_FOLDER --write_coco_json ${JSON_FOLDER}1_3.json --no_display --render_pose 0 --scale_number 3 --scale_gap 0.25 --frame_last 3558

 #     # 4 scales
-# $OP_BIN --num_gpu 1 --image_dir $IMAGE_FOLDER --write_coco_json ${JSON_FOLDER}1_4.json --no_display --render_pose 0 --num_gpu 1 --scale_number 4 --scale_gap 0.25 --net_resolution "1312x736" --frame_last 3558 --output_resolution "1280x720"
+# $OP_BIN --num_gpu 1 --image_dir $IMAGE_FOLDER --write_coco_json ${JSON_FOLDER}1_4.json --no_display --render_pose 0 --num_gpu 1 --scale_number 4 --scale_gap 0.25 --net_resolution "1312x736" --frame_last 3558

 # Debugging - Rendered frames saved
 # $OP_BIN --image_dir $IMAGE_FOLDER --write_images ${JSON_FOLDER}frameOutput --no_display
--- a/include/openpose/core/datum.hpp
+++ b/include/openpose/core/datum.hpp
@@ -59,11 +59,20 @@ namespace op
        /**
         * Body pose (x,y,score) locations for each person in the image.
         * It has been resized to the desired output resolution (e.g. `resolution` flag in the demo).
-         * If outputData is empty, then cvOutputData will also be empty.
         * Size: #people x #body parts (e.g. 18 for COCO or 15 for MPI) x 3 ((x,y) coordinates + score)
         */
        Array<float> poseKeypoints;

+        /**
+         * Body pose global confidence/score for each person in the image.
+         * It does not only consider the score of each body keypoint, but also the score of each PAF association.
+         * Optimized for COCO evaluation metric.
+         * It will highly penalyze people with missing body parts (e.g. cropped people on the borders of the image).
+         * If poseKeypoints is empty, then poseScores will also be empty.
+         * Size: #people
+         */
+        Array<float> poseScores;
+
        /**
         * Body pose heatmaps (body parts, background and/or PAFs) for the whole image.
         * This parameters is by default empty and disabled for performance. Each group (body parts, background and

--- a/include/openpose/core/resizeAndMergeCaffe.hpp
+++ b/include/openpose/core/resizeAndMergeCaffe.hpp
@@ -24,7 +24,7 @@ namespace op
        virtual void LayerSetUp(const std::vector<caffe::Blob<T>*>& bottom, const std::vector<caffe::Blob<T>*>& top);

        virtual void Reshape(const std::vector<caffe::Blob<T>*>& bottom, const std::vector<caffe::Blob<T>*>& top,
-                             const float factor, const bool mergeFirstDimension = true);
+                             const float netFactor, const float scaleFactor, const bool mergeFirstDimension = true);

        virtual inline const char* type() const { return "ResizeAndMerge"; }


--- a/include/openpose/filestream/cocoJsonSaver.hpp
+++ b/include/openpose/filestream/cocoJsonSaver.hpp
@@ -7,20 +7,23 @@
 namespace op
 {
    /**
-     *  The CocoJsonSaver class creates a COCO validation json file with details about the processed images. It inherits from Recorder.
+     *  The CocoJsonSaver class creates a COCO validation json file with details about the processed images. It
+     * inherits from Recorder.
     */
    class OP_API CocoJsonSaver
    {
    public:
        /**
-         * This constructor of CocoJsonSaver extends the Recorder::Recorder(const std::string & filePathToSave) constructor.
-         * @param filePathToSave const std::string parameter with the final file path where the generated json file will be saved.
+         * This constructor of CocoJsonSaver extends the Recorder::Recorder(const std::string & filePathToSave)
+         * constructor.
+         * @param filePathToSave const std::string parameter with the final file path where the generated json file
+         * will be saved.
         */
        explicit CocoJsonSaver(const std::string& filePathToSave, const bool humanReadable = true);

        ~CocoJsonSaver();

-        void record(const Array<float>& poseKeypoints, const std::string& imageName);
+        void record(const Array<float>& poseKeypoints, const Array<float>& poseScores, const std::string& imageName);

    private:
        JsonOfstream mJsonOfstream;

--- a/include/openpose/filestream/wCocoJsonSaver.hpp
+++ b/include/openpose/filestream/wCocoJsonSaver.hpp
@@ -60,7 +60,7 @@ namespace op
                // T* to T
                const auto& tDatum = tDatums->at(0);
                // Record json in COCO format
-                spCocoJsonSaver->record(tDatum.poseKeypoints, tDatum.name);
+                spCocoJsonSaver->record(tDatum.poseKeypoints, tDatum.poseScores, tDatum.name);
                // Profiling speed
                Profiler::timerEnd(profilerKey);
                Profiler::printAveragedTimeMsOnIterationX(profilerKey, __LINE__, __FUNCTION__, __FILE__);

--- a/include/openpose/pose/bodyPartConnectorBase.hpp
+++ b/include/openpose/pose/bodyPartConnectorBase.hpp
@@ -7,15 +7,15 @@
 namespace op
 {
    template <typename T>
-    OP_API void connectBodyPartsCpu(Array<T>& poseKeypoints, const T* const heatMapPtr, const T* const peaksPtr,
-                                    const PoseModel poseModel, const Point<int>& heatMapSize, const int maxPeaks,
-                                    const int interMinAboveThreshold, const T interThreshold, const int minSubsetCnt,
-                                    const T minSubsetScore, const T scaleFactor = 1.f);
+    OP_API void connectBodyPartsCpu(Array<T>& poseKeypoints, Array<T>& poseScores, const T* const heatMapPtr,
+                                    const T* const peaksPtr, const PoseModel poseModel, const Point<int>& heatMapSize,
+                                    const int maxPeaks, const T interMinAboveThreshold, const T interThreshold,
+                                    const int minSubsetCnt, const T minSubsetScore, const T scaleFactor = 1.f);

    template <typename T>
    OP_API void connectBodyPartsGpu(Array<T>& poseKeypoints, T* posePtr, const T* const heatMapPtr,
                                    const T* const peaksPtr, const PoseModel poseModel, const Point<int>& heatMapSize,
-                                    const int maxPeaks, const int interMinAboveThreshold, const T interThreshold,
+                                    const int maxPeaks, const T interMinAboveThreshold, const T interThreshold,
                                    const int minSubsetCnt, const T minSubsetScore, const T scaleFactor = 1.f);
 }


--- a/include/openpose/pose/bodyPartConnectorCaffe.hpp
+++ b/include/openpose/pose/bodyPartConnectorCaffe.hpp
@@ -30,7 +30,7 @@ namespace op

        void setPoseModel(const PoseModel poseModel);

-        void setInterMinAboveThreshold(const int interMinAboveThreshold);
+        void setInterMinAboveThreshold(const T interMinAboveThreshold);

        void setInterThreshold(const T interThreshold);

@@ -40,7 +40,8 @@ namespace op

        void setScaleNetToOutput(const T scaleNetToOutput);

-        virtual void Forward_cpu(const std::vector<caffe::Blob<T>*>& bottom, Array<T>& poseKeypoints);
+        virtual void Forward_cpu(const std::vector<caffe::Blob<T>*>& bottom, Array<T>& poseKeypoints,
+                                 Array<T>& poseScores);

        virtual void Forward_gpu(const std::vector<caffe::Blob<T>*>& bottom, const std::vector<caffe::Blob<T>*>& top,
                                 Array<T>& poseKeypoints);
@@ -53,7 +54,7 @@ namespace op

    private:
        PoseModel mPoseModel;
-        int mInterMinAboveThreshold;
+        T mInterMinAboveThreshold;
        T mInterThreshold;
        int mMinSubsetCnt;
        T mMinSubsetScore;

--- a/include/openpose/pose/poseExtractor.hpp
+++ b/include/openpose/pose/poseExtractor.hpp
@@ -35,6 +35,8 @@ namespace op

        Array<float> getPoseKeypoints() const;

+        Array<float> getPoseScores() const;
+
        float getScaleNetToOutput() const;

        double get(const PoseProperty property) const;
@@ -47,6 +49,7 @@ namespace op
        const PoseModel mPoseModel;
        Point<int> mNetOutputSize;
        Array<float> mPoseKeypoints;
+        Array<float> mPoseScores;
        float mScaleNetToOutput;

        void checkThread() const;

--- a/include/openpose/pose/poseParameters.hpp
+++ b/include/openpose/pose/poseParameters.hpp
@@ -287,8 +287,8 @@ namespace op
    const std::array<float, (int)PoseModel::Size>           POSE_DEFAULT_NMS_THRESHOLD{
        0.05f,      0.6f,       0.3f,       0.05f,      0.05f,      0.05f
    };
-    const std::array<unsigned int, (int)PoseModel::Size>    POSE_DEFAULT_CONNECT_INTER_MIN_ABOVE_THRESHOLD{
-        9,          8,          8,          9,          9,          9
+    const std::array<float, (int)PoseModel::Size>    POSE_DEFAULT_CONNECT_INTER_MIN_ABOVE_THRESHOLD{
+        0.95f,      0.95f,      0.95f,      0.95f,      0.95f,      0.95f
    };
    const std::array<float, (int)PoseModel::Size>           POSE_DEFAULT_CONNECT_INTER_THRESHOLD{
        0.05f,      0.01f,      0.01f,      0.05f,      0.05f,      0.05f

--- a/include/openpose/pose/wPoseExtractor.hpp
+++ b/include/openpose/pose/wPoseExtractor.hpp
@@ -63,6 +63,7 @@ namespace op
                                                 tDatum.scaleInputToNetInputs);
                    tDatum.poseHeatMaps = spPoseExtractor->getHeatMaps().clone();
                    tDatum.poseKeypoints = spPoseExtractor->getPoseKeypoints().clone();
+                    tDatum.poseScores = spPoseExtractor->getPoseScores().clone();
                    tDatum.scaleNetToOutput = spPoseExtractor->getScaleNetToOutput();
                }
                // Profiling speed

--- a/include/openpose/utilities/headers.hpp
+++ b/include/openpose/utilities/headers.hpp
@@ -13,6 +13,7 @@
 #include <openpose/utilities/openCv.hpp>
 #include <openpose/utilities/pointerContainer.hpp>
 #include <openpose/utilities/profiler.hpp>
+#include <openpose/utilities/standard.hpp>
 #include <openpose/utilities/string.hpp>

 #endif // OPENPOSE_UTILITIES_HEADERS_HPP
--- a/include/openpose/utilities/keypoint.hpp
+++ b/include/openpose/utilities/keypoint.hpp
@@ -21,11 +21,13 @@ namespace op
                                   const float thicknessCircleRatio, const float thicknessLineRatioWRTCircle,
                                   const float threshold);

-    OP_API Rectangle<float> getKeypointsRectangle(const Array<float>& keypoints, const int person, const int numberKeypoints, const float threshold);
+    OP_API Rectangle<float> getKeypointsRectangle(const Array<float>& keypoints, const int person,
+                                                  const int numberKeypoints, const float threshold);

    OP_API float getAverageScore(const Array<float>& keypoints, const int person);

-    OP_API float getKeypointsArea(const Array<float>& keypoints, const int person, const int numberKeypoints, const float threshold);
+    OP_API float getKeypointsArea(const Array<float>& keypoints, const int person, const int numberKeypoints,
+                                  const float threshold);

    OP_API int getBiggestPerson(const Array<float>& keypoints, const float threshold);
 }

--- a/include/openpose/utilities/standard.hpp
+++ b/include/openpose/utilities/standard.hpp
+#ifndef OPENPOSE_UTILITIES_STANDARD_HPP
+#define OPENPOSE_UTILITIES_STANDARD_HPP
+
+#include <openpose/core/common.hpp>
+
+namespace op
+{
+    template <typename T>
+    bool vectorsAreEqual(const std::vector<T>& vectorA, const std::vector<T>& vectorB)
+    {
+        try
+        {
+            if (vectorA.size() != vectorB.size())
+                return false;
+            else
+            {
+                for (auto i = 0u ; i < vectorA.size() ; i++)
+                    if (vectorA[i] != vectorB[i])
+                        return false;
+                return true;
+            }
+        }
+        catch (const std::exception& e)
+        {
+            error(e.what(), __LINE__, __FUNCTION__, __FILE__);
+            return false;
+        }
+    }
+}
+
+#endif // OPENPOSE_UTILITIES_STANDARD_HPP
--- a/include/openpose/wrapper/wrapper.hpp
+++ b/include/openpose/wrapper/wrapper.hpp
@@ -647,6 +647,7 @@ namespace op
                    if (wrapperStructPose.renderMode == RenderMode::Cpu)
                    {
                        poseCpuRenderer = std::make_shared<PoseCpuRenderer>(wrapperStructPose.poseModel,
+                                                                            wrapperStructPose.renderThreshold,
                                                                            wrapperStructPose.blendOriginalFrame);
                        cpuRenderers.emplace_back(std::make_shared<WPoseRenderer<TDatumsPtr>>(poseCpuRenderer));
                    }

--- a/src/openpose/core/datum.cpp
+++ b/src/openpose/core/datum.cpp
@@ -19,6 +19,7 @@ namespace op
        cvOutputData{datum.cvOutputData},
        // Resulting Array<float> data
        poseKeypoints{datum.poseKeypoints},
+        poseScores{datum.poseScores},
        poseHeatMaps{datum.poseHeatMaps},
        faceRectangles{datum.faceRectangles},
        faceKeypoints{datum.faceKeypoints},
@@ -48,6 +49,7 @@ namespace op
            cvOutputData = datum.cvOutputData;
            // Resulting Array<float> data
            poseKeypoints = datum.poseKeypoints;
+            poseScores = datum.poseScores,
            poseHeatMaps = datum.poseHeatMaps,
            faceRectangles = datum.faceRectangles,
            faceKeypoints = datum.faceKeypoints,
@@ -88,6 +90,7 @@ namespace op
            std::swap(cvOutputData, datum.cvOutputData);
            // Resulting Array<float> data
            std::swap(poseKeypoints, datum.poseKeypoints);
+            std::swap(poseScores, datum.poseScores);
            std::swap(poseHeatMaps, datum.poseHeatMaps);
            std::swap(faceRectangles, datum.faceRectangles);
            std::swap(faceKeypoints, datum.faceKeypoints);
@@ -119,6 +122,7 @@ namespace op
            std::swap(cvOutputData, datum.cvOutputData);
            // Resulting Array<float> data
            std::swap(poseKeypoints, datum.poseKeypoints);
+            std::swap(poseScores, datum.poseScores);
            std::swap(poseHeatMaps, datum.poseHeatMaps);
            std::swap(faceRectangles, datum.faceRectangles);
            std::swap(faceKeypoints, datum.faceKeypoints);
@@ -158,6 +162,7 @@ namespace op
            datum.cvOutputData = cvOutputData.clone();
            // Resulting Array<float> data
            datum.poseKeypoints = poseKeypoints.clone();
+            datum.poseScores = poseScores.clone();
            datum.poseHeatMaps = poseHeatMaps.clone();
            datum.faceRectangles = faceRectangles;
            datum.faceKeypoints = faceKeypoints.clone();

--- a/src/openpose/core/netCaffe.cpp
+++ b/src/openpose/core/netCaffe.cpp
@@ -7,6 +7,7 @@
 #endif
 #include <openpose/utilities/cuda.hpp>
 #include <openpose/utilities/fileSystem.hpp>
+#include <openpose/utilities/standard.hpp>
 #include <openpose/core/netCaffe.hpp>

 namespace op
@@ -71,20 +72,6 @@ namespace op
                error(e.what(), __LINE__, __FUNCTION__, __FILE__);
            }
        }
-
-        inline bool requiredReshapeNetCaffe(const std::vector<int>& dimensionsA, const std::vector<int>& dimensionsB)
-        {
-            try
-            {
-                return (dimensionsA[0] != dimensionsB[0] || dimensionsA[1] != dimensionsB[1]
-                        || dimensionsA[2] != dimensionsB[2] || dimensionsA[3] != dimensionsB[3]);
-            }
-            catch (const std::exception& e)
-            {
-                error(e.what(), __LINE__, __FUNCTION__, __FILE__);
-                return false;
-            }
-        }
    #endif

    NetCaffe::NetCaffe(const std::string& caffeProto, const std::string& caffeTrainedModel, const int gpuId,
@@ -157,7 +144,7 @@ namespace op
                    error("The Array inputData must have 4 dimensions: [batch size, 3 (RGB), height, width].",
                          __LINE__, __FUNCTION__, __FILE__);
                // Reshape Caffe net if required
-                if (requiredReshapeNetCaffe(upImpl->mNetInputSize4D, inputData.getSize()))
+                if (!vectorsAreEqual(upImpl->mNetInputSize4D, inputData.getSize()))
                {
                    upImpl->mNetInputSize4D = inputData.getSize();
                    reshapeNetCaffe(upImpl->upCaffeNet.get(), inputData.getSize());

--- a/src/openpose/core/resizeAndMergeCaffe.cpp
+++ b/src/openpose/core/resizeAndMergeCaffe.cpp
@@ -48,22 +48,27 @@ namespace op

    template <typename T>
    void ResizeAndMergeCaffe<T>::Reshape(const std::vector<caffe::Blob<T>*>& bottom,
-                                         const std::vector<caffe::Blob<T>*>& top, const float factor,
+                                         const std::vector<caffe::Blob<T>*>& top,
+                                         const float netFactor,
+                                         const float scaleFactor,
                                         const bool mergeFirstDimension)
    {
        try
        {
            #ifdef USE_CAFFE
-                auto bottomBlob = bottom.at(0);
-                auto topBlob = top.at(0);
-
-                // Top shape
+                // Data
+                const auto* bottomBlob = bottom.at(0);
+                auto* topBlob = top.at(0);
+                // Set top shape
                auto topShape = bottomBlob->shape();
                topShape[0] = (mergeFirstDimension ? 1 : bottomBlob->shape(0));
-                topShape[2] = intRound(topShape[2] * factor);
-                topShape[3] = intRound(topShape[3] * factor);
+                // -1 and later +1 to take into account that we are using 0-based index
+                // E.g. 100x100 image --> 200x200 --> 0-99 to 0-199 --> scale = 199/99 (not 2!)
+                // E.g. 101x101 image --> 201x201 --> scale = 2
+                // Test: pixel 0 --> 0, pixel 99 (ex 1) --> 199, pixel 100 (ex 2) --> 200
+                topShape[2] = intRound((topShape[2]*netFactor - 1.f) * scaleFactor + 1);
+                topShape[3] = intRound((topShape[3]*netFactor - 1.f) * scaleFactor + 1);
                topBlob->Reshape(topShape);
-
                // Array sizes
                mTopSize = std::array<int, 4>{topBlob->shape(0), topBlob->shape(1), topBlob->shape(2),
                                              topBlob->shape(3)};

--- a/src/openpose/face/faceDetectorOpenCV.cpp
+++ b/src/openpose/face/faceDetectorOpenCV.cpp
@@ -25,11 +25,11 @@ namespace op
            // Image to grey and pyrDown
            cv::Mat frameGray;
            cv::cvtColor(cvInputData, frameGray, cv::COLOR_BGR2GRAY);
-            auto multiplier = 1;
+            auto multiplier = 1.f;
            while (frameGray.cols * frameGray.rows > 640*360)
            {
                cv::pyrDown(frameGray, frameGray);
-                multiplier *= 2;
+                multiplier *= 2.f;
            }
            // Face detection - Example from:
            // http://docs.opencv.org/2.4/doc/tutorials/objdetect/cascade_classifier/cascade_classifier.html
@@ -40,10 +40,10 @@ namespace op
            for(auto i = 0u; i < detectedFaces.size(); i++)
            {
                // Enlarge detected rectangle by 1.5x, so that it covers the whole face
-                faceRectangles.at(i).x = detectedFaces.at(i).x - 0.25*detectedFaces.at(i).width;
-                faceRectangles.at(i).y = detectedFaces.at(i).y - 0.25*detectedFaces.at(i).height;
-                faceRectangles.at(i).width = 1.5*detectedFaces.at(i).width;
-                faceRectangles.at(i).height = 1.5*detectedFaces.at(i).height;
+                faceRectangles.at(i).x = detectedFaces.at(i).x - 0.25f*detectedFaces.at(i).width;
+                faceRectangles.at(i).y = detectedFaces.at(i).y - 0.25f*detectedFaces.at(i).height;
+                faceRectangles.at(i).width = 1.5f*detectedFaces.at(i).width;
+                faceRectangles.at(i).height = 1.5f*detectedFaces.at(i).height;
                faceRectangles.at(i) *= multiplier;
            }
            return faceRectangles;

--- a/src/openpose/face/faceExtractorCaffe.cpp
+++ b/src/openpose/face/faceExtractorCaffe.cpp
@@ -80,7 +80,7 @@ namespace op
                // HeatMaps extractor blob and layer
                const bool mergeFirstDimension = true;
                resizeAndMergeCaffe->Reshape({caffeNetOutputBlob.get()}, {heatMapsBlob.get()},
-                                             FACE_CCN_DECREASE_FACTOR, mergeFirstDimension);
+                                             FACE_CCN_DECREASE_FACTOR, 1.f, mergeFirstDimension);
                // Pose extractor blob and layer
                maximumCaffe->Reshape({heatMapsBlob.get()}, {peaksBlob.get()});
                // Cuda check

--- a/src/openpose/filestream/cocoJsonSaver.cpp
+++ b/src/openpose/filestream/cocoJsonSaver.cpp
@@ -28,10 +28,14 @@ namespace op
        }
    }

-    void CocoJsonSaver::record(const Array<float>& poseKeypoints, const std::string& imageName)
+    void CocoJsonSaver::record(const Array<float>& poseKeypoints, const Array<float>& poseScores,
+                               const std::string& imageName)
    {
        try
        {
+            // Security checks
+            if ((size_t)poseKeypoints.getSize(0) != poseScores.getVolume())
+                error("Dimension mismatch between poseKeypoints and poseScores.", __LINE__, __FUNCTION__, __FILE__);
            const auto numberPeople = poseKeypoints.getSize(0);
            const auto numberBodyParts = poseKeypoints.getSize(1);
            const auto imageId = getLastNumber(imageName);
@@ -75,11 +79,12 @@ namespace op
                for (auto bodyPart = 0u ; bodyPart < indexesInCocoOrder.size() ; bodyPart++)
                {
                    const auto finalIndex = 3*(person*numberBodyParts + indexesInCocoOrder.at(bodyPart));
-                    mJsonOfstream.plainText(poseKeypoints[finalIndex] + 0.5f);
+                    mJsonOfstream.plainText(poseKeypoints[finalIndex]);
                    mJsonOfstream.comma();
-                    mJsonOfstream.plainText(poseKeypoints[finalIndex+1] + 0.5f);
+                    mJsonOfstream.plainText(poseKeypoints[finalIndex+1]);
                    mJsonOfstream.comma();
                    mJsonOfstream.plainText((poseKeypoints[finalIndex+2] > 0.f ? 1 : 0));
+                    // mJsonOfstream.plainText(poseKeypoints[finalIndex+2]); // For debugging
                    if (bodyPart < indexesInCocoOrder.size() - 1u)
                        mJsonOfstream.comma();
                }
@@ -88,7 +93,7 @@ namespace op

                // score
                mJsonOfstream.key("score");
-                mJsonOfstream.plainText("0.0");
+                mJsonOfstream.plainText(poseScores[person]);

                mJsonOfstream.objectClose();
            }

--- a/src/openpose/gui/gui.cpp
+++ b/src/openpose/gui/gui.cpp
@@ -97,7 +97,7 @@ namespace op
                        poseExtractor->increase(PoseProperty::ConnectInterThreshold, 0.005f * (castedKey=='[' ? -1 : 1));
                else if (castedKey=='{' || castedKey=='}')
                    for (auto& poseExtractor : poseExtractors)
-                        poseExtractor->increase(PoseProperty::ConnectInterMinAboveThreshold, (castedKey=='{' ? -1 : 1));
+                        poseExtractor->increase(PoseProperty::ConnectInterMinAboveThreshold, (castedKey=='{' ? -0.1f : 0.1f));
                else if (castedKey==';' || castedKey=='\'')
                    for (auto& poseExtractor : poseExtractors)
                        poseExtractor->increase(PoseProperty::ConnectMinSubsetCnt, (castedKey==';' ? -1 : 1));

--- a/src/openpose/hand/handExtractorCaffe.cpp
+++ b/src/openpose/hand/handExtractorCaffe.cpp
@@ -166,7 +166,7 @@ namespace op
                // HeatMaps extractor blob and layer
                const bool mergeFirstDimension = true;
                resizeAndMergeCaffe->Reshape({caffeNetOutputBlob.get()}, {heatMapsBlob.get()},
-                                             HAND_CCN_DECREASE_FACTOR, mergeFirstDimension);
+                                             HAND_CCN_DECREASE_FACTOR, 1.f, mergeFirstDimension);
                // Pose extractor blob and layer
                maximumCaffe->Reshape({heatMapsBlob.get()}, {peaksBlob.get()});
                // Cuda check

--- a/src/openpose/pose/bodyPartConnectorBase.cpp
+++ b/src/openpose/pose/bodyPartConnectorBase.cpp
@@ -6,10 +6,10 @@
 namespace op
 {
    template <typename T>
-    void connectBodyPartsCpu(Array<T>& poseKeypoints, const T* const heatMapPtr, const T* const peaksPtr,
-                             const PoseModel poseModel, const Point<int>& heatMapSize, const int maxPeaks,
-                             const int interMinAboveThreshold, const T interThreshold, const int minSubsetCnt,
-                             const T minSubsetScore, const T scaleFactor)
+    void connectBodyPartsCpu(Array<T>& poseKeypoints, Array<T>& poseScores, const T* const heatMapPtr,
+                             const T* const peaksPtr, const PoseModel poseModel, const Point<int>& heatMapSize,
+                             const int maxPeaks, const T interMinAboveThreshold, const T interThreshold,
+                             const int minSubsetCnt, const T minSubsetScore, const T scaleFactor)
    {
        try
        {
@@ -36,7 +36,7 @@ namespace op
                const auto nA = intRound(candidateA[0]);
                const auto nB = intRound(candidateB[0]);

-                // add parts into the subset in special case
+                // Add parts into the subset in special case
                if (nA == 0 || nB == 0)
                {
                    // Change w.r.t. other
@@ -142,7 +142,6 @@ namespace op
                else // if (nA != 0 && nB != 0)
                {
                    std::vector<std::tuple<double, int, int>> temp;
-                    const auto numInter = 10;
                    const auto* const mapX = heatMapPtr + mapIdx[2*pairIndex] * heatMapOffset;
                    const auto* const mapY = heatMapPtr + mapIdx[2*pairIndex+1] * heatMapOffset;
                    for (auto i = 1; i <= nA; i++)
@@ -151,6 +150,8 @@ namespace op
                        {
                            const auto dX = candidateB[j*3] - candidateA[i*3];
                            const auto dY = candidateB[j*3+1] - candidateA[i*3+1];
+                            const auto dMax = fastMax(std::abs(dX), std::abs(dY));
+                            const auto numberPointsInLine = fastMax(5, fastMin(25, intRound(std::sqrt(5*dMax))));
                            const auto normVec = T(std::sqrt( dX*dX + dY*dY ));
                            // If the peaksPtr are coincident. Don't connect them.
                            if (normVec > 1e-6)
@@ -162,10 +163,12 @@ namespace op

                                auto sum = 0.;
                                auto count = 0;
-                                for (auto lm=0; lm < numInter; lm++)
+                                const auto dXInLine = dX/numberPointsInLine;
+                                const auto dYInLine = dY/numberPointsInLine;
+                                for (auto lm = 0; lm < numberPointsInLine; lm++)
                                {
-                                    const auto mX = fastMin(heatMapSize.x-1, intRound(sX + lm*dX/numInter));
-                                    const auto mY = fastMin(heatMapSize.y-1, intRound(sY + lm*dY/numInter));
+                                    const auto mX = fastMin(heatMapSize.x-1, intRound(sX + lm*dXInLine));
+                                    const auto mY = fastMin(heatMapSize.y-1, intRound(sY + lm*dYInLine));
                                    checkGE(mX, 0, "", __LINE__, __FUNCTION__, __FILE__);
                                    checkGE(mY, 0, "", __LINE__, __FUNCTION__, __FILE__);
                                    const auto idx = mY * heatMapSize.x + mX;
@@ -178,7 +181,7 @@ namespace op
                                }

                                // parts score + connection score
-                                if (count > interMinAboveThreshold)
+                                if (count/(float)numberPointsInLine > interMinAboveThreshold)
                                    temp.emplace_back(std::make_tuple(sum/count, i, j));
                            }
                        }
@@ -190,7 +193,6 @@ namespace op
                        std::sort(temp.begin(), temp.end(), std::greater<std::tuple<T, int, int>>());

                    std::vector<std::tuple<int, int, double>> connectionK;
-
                    const auto minAB = fastMin(nA, nB);
                    std::vector<int> occurA(nA, 0);
                    std::vector<int> occurB(nB, 0);
@@ -214,63 +216,63 @@ namespace op
                    }

                    // Cluster all the body part candidates into subset based on the part connection
-                    // initialize first body part connection 15&16
-                    if (pairIndex==0)
+                    if (!connectionK.empty())
                    {
-                        for (const auto connectionKI : connectionK)
+                        // initialize first body part connection 15&16
+                        if (pairIndex==0)
                        {
-                            std::vector<int> rowVector(numberBodyParts+3, 0);
-                            const auto indexA = std::get<0>(connectionKI);
-                            const auto indexB = std::get<1>(connectionKI);
-                            const auto score = std::get<2>(connectionKI);
-                            rowVector[bodyPartPairs[0]] = indexA;
-                            rowVector[bodyPartPairs[1]] = indexB;
-                            rowVector[subsetCounterIndex] = 2;
-                            // add the score of parts and the connection
-                            const auto subsetScore = peaksPtr[indexA] + peaksPtr[indexB] + score;
-                            subset.emplace_back(std::make_pair(rowVector, subsetScore));
+                            for (const auto connectionKI : connectionK)
+                            {
+                                std::vector<int> rowVector(numberBodyParts+3, 0);
+                                const auto indexA = std::get<0>(connectionKI);
+                                const auto indexB = std::get<1>(connectionKI);
+                                const auto score = std::get<2>(connectionKI);
+                                rowVector[bodyPartPairs[0]] = indexA;
+                                rowVector[bodyPartPairs[1]] = indexB;
+                                rowVector[subsetCounterIndex] = 2;
+                                // add the score of parts and the connection
+                                const auto subsetScore = peaksPtr[indexA] + peaksPtr[indexB] + score;
+                                subset.emplace_back(std::make_pair(rowVector, subsetScore));
+                            }
                        }
-                    }
-                    // Add ears connections (in case person is looking to opposite direction to camera)
-                    else if (((poseModel == PoseModel::COCO_18
-                                || poseModel == PoseModel::BODY_18) && (pairIndex==17 || pairIndex==18))
-                             || (poseModel == PoseModel::BODY_19 && (pairIndex==18 || pairIndex==19))
-                             || (poseModel == PoseModel::BODY_23 && (pairIndex==22 || pairIndex==23)))
-                    {
-                        for (const auto& connectionKI : connectionK)
+                        // Add ears connections (in case person is looking to opposite direction to camera)
+                        else if (((poseModel == PoseModel::COCO_18
+                                    || poseModel == PoseModel::BODY_18) && (pairIndex==17 || pairIndex==18))
+                                 || (poseModel == PoseModel::BODY_19 && (pairIndex==18 || pairIndex==19))
+                                 || (poseModel == PoseModel::BODY_23 && (pairIndex==22 || pairIndex==23)))
                        {
-                            const auto indexA = std::get<0>(connectionKI);
-                            const auto indexB = std::get<1>(connectionKI);
-                            for (auto& subsetJ : subset)
+                            for (const auto& connectionKI : connectionK)
                            {
-                                auto& subsetJFirst = subsetJ.first[bodyPartA];
-                                auto& subsetJFirstPlus1 = subsetJ.first[bodyPartB];
-                                if (subsetJFirst == indexA && subsetJFirstPlus1 == 0)
-                                    subsetJFirstPlus1 = indexB;
-                                else if (subsetJFirstPlus1 == indexB && subsetJFirst == 0)
-                                    subsetJFirst = indexA;
+                                const auto indexA = std::get<0>(connectionKI);
+                                const auto indexB = std::get<1>(connectionKI);
+                                for (auto& subsetJ : subset)
+                                {
+                                    auto& subsetJFirst = subsetJ.first[bodyPartA];
+                                    auto& subsetJFirstPlus1 = subsetJ.first[bodyPartB];
+                                    if (subsetJFirst == indexA && subsetJFirstPlus1 == 0)
+                                        subsetJFirstPlus1 = indexB;
+                                    else if (subsetJFirstPlus1 == indexB && subsetJFirst == 0)
+                                        subsetJFirst = indexA;
+                                }
                            }
                        }
-                    }
-                    else
-                    {
-                        if (!connectionK.empty())
+                        else
                        {
                            // A is already in the subset, find its connection B
-                            for (auto i = 0u; i < connectionK.size(); i++)
+                            for (const auto& connectionKI : connectionK)
                            {
-                                const auto indexA = std::get<0>(connectionK[i]);
-                                const auto indexB = std::get<1>(connectionK[i]);
-                                const auto score = std::get<2>(connectionK[i]);
+                                const auto indexA = std::get<0>(connectionKI);
+                                const auto indexB = std::get<1>(connectionKI);
+                                const auto score = std::get<2>(connectionKI);
                                auto num = 0;
-                                for (auto j = 0u; j < subset.size(); j++)
+                                for (auto& subsetJ : subset)
                                {
-                                    if (subset[j].first[bodyPartA] == indexA)
+                                    if (subsetJ.first[bodyPartA] == indexA)
                                    {
-                                        subset[j].first[bodyPartB] = indexB;
+                                        subsetJ.first[bodyPartB] = indexB;
                                        num++;
-                                        subset[j].first[subsetCounterIndex] = subset[j].first[subsetCounterIndex] + 1;
-                                        subset[j].second = subset[j].second + peaksPtr[indexB] + score;
+                                        subsetJ.first[subsetCounterIndex] = subsetJ.first[subsetCounterIndex] + 1;
+                                        subsetJ.second += peaksPtr[indexB] + score;
                                    }
                                }
                                // if can not find partA in the subset, create a new subset
@@ -300,7 +302,7 @@ namespace op
            {
                const auto subsetCounter = subset[index].first[subsetCounterIndex];
                const auto subsetScore = subset[index].second;
-                if (subsetCounter >= minSubsetCnt && (subsetScore/subsetCounter) > minSubsetScore)
+                if (subsetCounter >= minSubsetCnt && (subsetScore/subsetCounter) >= minSubsetScore)
                {
                    numberPeople++;
                    validSubsetIndexes.emplace_back(index);
@@ -314,20 +316,30 @@ namespace op

            // Fill and return poseKeypoints
            if (numberPeople > 0)
+            {
                poseKeypoints.reset({numberPeople, (int)numberBodyParts, 3});
+                poseScores.reset(numberPeople);
+            }
            else
+            {
                poseKeypoints.reset();
+                poseScores.reset();
+            }
+            const auto numberBodyPartsAndPAFs = numberBodyParts + numberBodyPartPairs;
            for (auto person = 0u ; person < validSubsetIndexes.size() ; person++)
            {
-                const auto& subsetI = subset[validSubsetIndexes[person]].first;
+                const auto& subsetPair = subset[validSubsetIndexes[person]];
+                const auto& subsetI = subsetPair.first;
                for (auto bodyPart = 0u; bodyPart < numberBodyParts; bodyPart++)
                {
                    const auto baseOffset = (person*numberBodyParts + bodyPart) * 3;
                    const auto bodyPartIndex = subsetI[bodyPart];
                    if (bodyPartIndex > 0)
                    {
-                        poseKeypoints[baseOffset] = peaksPtr[bodyPartIndex-2] * scaleFactor;
-                        poseKeypoints[baseOffset + 1] = peaksPtr[bodyPartIndex-1] * scaleFactor;
+                        // Best results for 1 scale: x + 0, y + 0.5
+                        // +0.5 to both to keep Matlab format
+                        poseKeypoints[baseOffset] = peaksPtr[bodyPartIndex-2] * scaleFactor + 0.5f;
+                        poseKeypoints[baseOffset + 1] = peaksPtr[bodyPartIndex-1] * scaleFactor + 0.5f;
                        poseKeypoints[baseOffset + 2] = peaksPtr[bodyPartIndex];
                    }
                    else
@@ -337,6 +349,7 @@ namespace op
                        poseKeypoints[baseOffset + 2] = 0.f;
                    }
                }
+                poseScores[person] = subsetPair.second / (float)(numberBodyPartsAndPAFs);
            }
        }
        catch (const std::exception& e)
@@ -345,16 +358,16 @@ namespace op
        }
    }

-    template void connectBodyPartsCpu(Array<float>& poseKeypoints, const float* const heatMapPtr,
-                                      const float* const peaksPtr, const PoseModel poseModel,
-                                      const Point<int>& heatMapSize, const int maxPeaks,
-                                      const int interMinAboveThreshold, const float interThreshold,
-                                      const int minSubsetCnt, const float minSubsetScore,
-                                      const float scaleFactor);
-    template void connectBodyPartsCpu(Array<double>& poseKeypoints, const double* const heatMapPtr,
-                                      const double* const peaksPtr, const PoseModel poseModel,
-                                      const Point<int>& heatMapSize, const int maxPeaks,
-                                      const int interMinAboveThreshold, const double interThreshold,
-                                      const int minSubsetCnt, const double minSubsetScore,
-                                      const double scaleFactor);
+    template void connectBodyPartsCpu(Array<float>& poseKeypoints, Array<float>& poseScores,
+                                      const float* const heatMapPtr, const float* const peaksPtr,
+                                      const PoseModel poseModel, const Point<int>& heatMapSize,
+                                      const int maxPeaks, const float interMinAboveThreshold,
+                                      const float interThreshold, const int minSubsetCnt,
+                                      const float minSubsetScore, const float scaleFactor);
+    template void connectBodyPartsCpu(Array<double>& poseKeypoints, Array<double>& poseScores,
+                                      const double* const heatMapPtr, const double* const peaksPtr,
+                                      const PoseModel poseModel, const Point<int>& heatMapSize,
+                                      const int maxPeaks, const double interMinAboveThreshold,
+                                      const double interThreshold, const int minSubsetCnt,
+                                      const double minSubsetScore, const double scaleFactor);
 }
--- a/src/openpose/pose/bodyPartConnectorBase.cu
+++ b/src/openpose/pose/bodyPartConnectorBase.cu
@@ -6,7 +6,7 @@ namespace op
    template <typename T>
    void connectBodyPartsGpu(Array<T>& poseKeypoints, T* posePtr, const T* const heatMapPtr, const T* const peaksPtr,
                             const PoseModel poseModel, const Point<int>& heatMapSize, const int maxPeaks,
-                             const int interMinAboveThreshold, const T interThreshold, const int minSubsetCnt,
+                             const T interMinAboveThreshold, const T interThreshold, const int minSubsetCnt,
                             const T minSubsetScore, const T scaleFactor)
    {
        try
@@ -35,11 +35,11 @@ namespace op
    template void connectBodyPartsGpu(Array<float>& poseKeypoints, float* posePtr, const float* const heatMapPtr,
                                      const float* const peaksPtr, const PoseModel poseModel,
                                      const Point<int>& heatMapSize, const int maxPeaks,
-                                      const int interMinAboveThreshold, const float interThreshold,
+                                      const float interMinAboveThreshold, const float interThreshold,
                                      const int minSubsetCnt, const float minSubsetScore, const float scaleFactor);
    template void connectBodyPartsGpu(Array<double>& poseKeypoints, double* posePtr, const double* const heatMapPtr,
                                      const double* const peaksPtr, const PoseModel poseModel,
                                      const Point<int>& heatMapSize, const int maxPeaks,
-                                      const int interMinAboveThreshold, const double interThreshold,
+                                      const double interMinAboveThreshold, const double interThreshold,
                                      const int minSubsetCnt, const double minSubsetScore, const double scaleFactor);
 }
--- a/src/openpose/pose/bodyPartConnectorCaffe.cpp
+++ b/src/openpose/pose/bodyPartConnectorCaffe.cpp
@@ -92,7 +92,7 @@ namespace op
    }

    template <typename T>
-    void BodyPartConnectorCaffe<T>::setInterMinAboveThreshold(const int interMinAboveThreshold)
+    void BodyPartConnectorCaffe<T>::setInterMinAboveThreshold(const T interMinAboveThreshold)
    {
        try
        {
@@ -157,7 +157,8 @@ namespace op
    }

    template <typename T>
-    void BodyPartConnectorCaffe<T>::Forward_cpu(const std::vector<caffe::Blob<T>*>& bottom, Array<T>& poseKeypoints)
+    void BodyPartConnectorCaffe<T>::Forward_cpu(const std::vector<caffe::Blob<T>*>& bottom, Array<T>& poseKeypoints,
+                                                Array<T>& poseScores)
    {
        try
        {
@@ -166,7 +167,7 @@ namespace op
                const auto* const heatMapsPtr = heatMapsBlob->cpu_data();                       // ~8.5ms / 114
                const auto* const peaksPtr = bottom.at(1)->cpu_data();                          // ~0.02ms
                const auto maxPeaks = mTopSize[1];
-                connectBodyPartsCpu(poseKeypoints, heatMapsPtr, peaksPtr, mPoseModel,
+                connectBodyPartsCpu(poseKeypoints, poseScores, heatMapsPtr, peaksPtr, mPoseModel,
                                    Point<int>{heatMapsBlob->shape(3), heatMapsBlob->shape(2)},
                                    maxPeaks, mInterMinAboveThreshold, mInterThreshold,
                                    mMinSubsetCnt, mMinSubsetScore, mScaleNetToOutput);

--- a/src/openpose/pose/poseExtractor.cpp
+++ b/src/openpose/pose/poseExtractor.cpp
@@ -46,7 +46,7 @@ namespace op
    PoseExtractor::PoseExtractor(const PoseModel poseModel, const std::vector<HeatMapType>& heatMapTypes,
                                 const ScaleMode heatMapScale) :
        mPoseModel{poseModel},
-        mNetOutputSize{0, 0},
+        mNetOutputSize{0,0},
        mHeatMapTypes{heatMapTypes},
        mHeatMapScaleMode{heatMapScale}
    {
@@ -103,15 +103,19 @@ namespace op
            Array<float> heatMaps;
            if (!mHeatMapTypes.empty())
            {
+                // Get heatmaps size
+                const auto heatMapSize = getHeatMapSize();
+
                // Allocate memory
                const auto numberHeatMapChannels = getNumberHeatMapChannels(mHeatMapTypes, mPoseModel);
-                heatMaps.reset({numberHeatMapChannels, mNetOutputSize.y, mNetOutputSize.x});
+                heatMaps.reset({numberHeatMapChannels, heatMapSize[2], heatMapSize[3]});

                // Copy memory
                const auto channelOffset = heatMaps.getVolume(1, 2);
                const auto volumeBodyParts = POSE_NUMBER_BODY_PARTS[(int)mPoseModel] * channelOffset;
                const auto volumePAFs = POSE_BODY_PART_PAIRS[(int)mPoseModel].size() * channelOffset;
                unsigned int totalOffset = 0u;
+                // Body parts
                if (heatMapTypesHas(mHeatMapTypes, HeatMapType::Parts))
                {
                    #ifdef USE_CUDA
@@ -135,6 +139,7 @@ namespace op
                            heatMaps[i] = fastTruncate(heatMaps[i]);
                    totalOffset += (unsigned int)volumeBodyParts;
                }
+                // Background
                if (heatMapTypesHas(mHeatMapTypes, HeatMapType::Background))
                {
                    #ifdef USE_CUDA
@@ -160,6 +165,7 @@ namespace op
                            heatMapsPtr[i] = fastTruncate(heatMapsPtr[i]);
                    totalOffset += (unsigned int)channelOffset;
                }
+                // PAFs
                if (heatMapTypesHas(mHeatMapTypes, HeatMapType::PAFs))
                {
                    #ifdef USE_CUDA
@@ -216,6 +222,20 @@ namespace op
        }
    }

+    Array<float> PoseExtractor::getPoseScores() const
+    {
+        try
+        {
+            checkThread();
+            return mPoseScores;
+        }
+        catch (const std::exception& e)
+        {
+            error(e.what(), __LINE__, __FUNCTION__, __FILE__);
+            return Array<float>{};
+        }
+    }
+
    float PoseExtractor::getScaleNetToOutput() const
    {
        try

--- a/src/openpose/pose/poseExtractorCaffe.cpp
+++ b/src/openpose/pose/poseExtractorCaffe.cpp
@@ -10,6 +10,7 @@
 #include <openpose/utilities/cuda.hpp>
 #include <openpose/utilities/fastMath.hpp>
 #include <openpose/utilities/openCv.hpp>
+#include <openpose/utilities/standard.hpp>
 #include <openpose/pose/poseExtractorCaffe.hpp>

 namespace op
@@ -17,11 +18,12 @@ namespace op
    struct PoseExtractorCaffe::ImplPoseExtractorCaffe
    {
        #ifdef USE_CAFFE
-            std::vector<int> mNetInputSize4D;
            std::shared_ptr<NetCaffe> spNetCaffe;
            std::shared_ptr<ResizeAndMergeCaffe<float>> spResizeAndMergeCaffe;
            std::shared_ptr<NmsCaffe<float>> spNmsCaffe;
            std::shared_ptr<BodyPartConnectorCaffe<float>> spBodyPartConnectorCaffe;
+            std::vector<int> mNetInputSize4D;
+            std::vector<double> mScaleInputToNetInputs;
            // Init with thread
            boost::shared_ptr<caffe::Blob<float>> spCaffeNetOutputBlob;
            std::shared_ptr<caffe::Blob<float>> spHeatMapsBlob;
@@ -30,7 +32,6 @@ namespace op

            ImplPoseExtractorCaffe(const PoseModel poseModel, const int gpuId,
                                   const std::string& modelFolder, const bool enableGoogleLogging) :
-                mNetInputSize4D{0,0,0,0},
                spNetCaffe{std::make_shared<NetCaffe>(modelFolder + POSE_PROTOTXT[(int)poseModel],
                                                      modelFolder + POSE_TRAINED_MODEL[(int)poseModel], gpuId,
                                                      enableGoogleLogging)},
@@ -56,9 +57,8 @@ namespace op
            try
            {
                // HeatMaps extractor blob and layer
-                UNUSED(scaleInputToNetInput);
                resizeAndMergeCaffe->Reshape({caffeNetOutputBlob.get()}, {heatMapsBlob.get()},
-                                             POSE_CCN_DECREASE_FACTOR[(int)poseModel]);
+                                             POSE_CCN_DECREASE_FACTOR[(int)poseModel], 1.f/scaleInputToNetInput);
                // Pose extractor blob and layer
                nmsCaffe->Reshape({heatMapsBlob.get()}, {peaksBlob.get()}, POSE_MAX_PEAKS[(int)poseModel]);
                // Pose extractor blob and layer
@@ -71,21 +71,6 @@ namespace op
                error(e.what(), __LINE__, __FUNCTION__, __FILE__);
            }
        }
-
-        inline bool requiredReshapePoseExtractorCaffe(const std::vector<int>& dimensionsA,
-                                                      const std::vector<int>& dimensionsB)
-        {
-            try
-            {
-                return (dimensionsA[0] != dimensionsB[0] || dimensionsA[1] != dimensionsB[1]
-                        || dimensionsA[2] != dimensionsB[2] || dimensionsA[3] != dimensionsB[3]);
-            }
-            catch (const std::exception& e)
-            {
-                error(e.what(), __LINE__, __FUNCTION__, __FILE__);
-                return false;
-            }
-        }
    #endif

    PoseExtractorCaffe::PoseExtractorCaffe(const PoseModel poseModel, const std::string& modelFolder,
@@ -158,17 +143,21 @@ namespace op
                    error("Empty inputNetData.", __LINE__, __FUNCTION__, __FILE__);

                // 1. Caffe deep network
-                upImpl->spNetCaffe->forwardPass(inputNetData);                                                  // ~80ms
+                upImpl->spNetCaffe->forwardPass(inputNetData);                                                 // ~80ms

                // Reshape blobs if required
-                if (requiredReshapePoseExtractorCaffe(upImpl->mNetInputSize4D, inputNetData.getSize()))
+                // Note: In order to resize to input size to have same results as Matlab, uncomment the commented lines
+                if (!vectorsAreEqual(upImpl->mNetInputSize4D, inputNetData.getSize()))
+                    // || !vectorsAreEqual(upImpl->mScaleInputToNetInputs, scaleInputToNetInputs))
                {
                    upImpl->mNetInputSize4D = inputNetData.getSize();
                    mNetOutputSize = Point<int>{upImpl->mNetInputSize4D[3], upImpl->mNetInputSize4D[2]};
+                    // upImpl->mScaleInputToNetInputs = scaleInputToNetInputs;
                    reshapePoseExtractorCaffe(upImpl->spResizeAndMergeCaffe, upImpl->spNmsCaffe,
                                              upImpl->spBodyPartConnectorCaffe, upImpl->spCaffeNetOutputBlob,
                                              upImpl->spHeatMapsBlob, upImpl->spPeaksBlob, upImpl->spPoseBlob,
-                                              scaleInputToNetInputs[0], mPoseModel);
+                                              1.f, mPoseModel);
+                                              // scaleInputToNetInputs[0], mPoseModel);
                }

                // 2. Resize heat maps + merge different scales
@@ -192,16 +181,18 @@ namespace op
                #endif

                // Get scale net to output (i.e. image input)
+                // Note: In order to resize to input size, (un)comment the following lines
                const auto scaleProducerToNetInput = resizeGetScaleFactor(inputDataSize, mNetOutputSize);
                const Point<int> netSize{intRound(scaleProducerToNetInput*inputDataSize.x),
                                         intRound(scaleProducerToNetInput*inputDataSize.y)};
                mScaleNetToOutput = {(float)resizeGetScaleFactor(netSize, inputDataSize)};
+                // mScaleNetToOutput = 1.f;

                // 4. Connecting body parts
+                // Get scale net to output (i.e. image input)
                upImpl->spBodyPartConnectorCaffe->setScaleNetToOutput(mScaleNetToOutput);
-                // upImpl->spBodyPartConnectorCaffe->setScaleNetToOutput(1);
                upImpl->spBodyPartConnectorCaffe->setInterMinAboveThreshold(
-                    (int)get(PoseProperty::ConnectInterMinAboveThreshold)
+                    (float)get(PoseProperty::ConnectInterMinAboveThreshold)
                );
                upImpl->spBodyPartConnectorCaffe->setInterThreshold((float)get(PoseProperty::ConnectInterThreshold));
                upImpl->spBodyPartConnectorCaffe->setMinSubsetCnt((int)get(PoseProperty::ConnectMinSubsetCnt));
@@ -215,7 +206,7 @@ namespace op
                // #else
                    upImpl->spBodyPartConnectorCaffe->Forward_cpu({upImpl->spHeatMapsBlob.get(),
                                                                   upImpl->spPeaksBlob.get()},
-                                                                  mPoseKeypoints);
+                                                                  mPoseKeypoints, mPoseScores);
                // #endif
            #else
                UNUSED(inputNetData);

--- a/src/openpose/pose/poseGpuRenderer.cpp
+++ b/src/openpose/pose/poseGpuRenderer.cpp
@@ -100,6 +100,9 @@ namespace op
                    }
                    else
                    {
+                        // If resized to input resolution: Replace scaleNetToOutput * scaleInputToOutput by
+                        // scaleInputToOutput, and comment the security checks.
+                        // Security checks
                        if (scaleNetToOutput == -1.f)
                            error("Non valid scaleNetToOutput.", __LINE__, __FUNCTION__, __FILE__);
                        // Parameters

--- a/src/openpose/utilities/openCv.cpp
+++ b/src/openpose/utilities/openCv.cpp
@@ -3,7 +3,8 @@

 namespace op
 {
-    void putTextOnCvMat(cv::Mat& cvMat, const std::string& textToDisplay, const Point<int>& position, const cv::Scalar& color, const bool normalizeWidth)
+    void putTextOnCvMat(cv::Mat& cvMat, const std::string& textToDisplay, const Point<int>& position,
+                        const cv::Scalar& color, const bool normalizeWidth)
    {
        try
        {
@@ -14,8 +15,11 @@ namespace op
            const auto shadowOffset = 2;
            int baseline = 0;
            const auto textSize = cv::getTextSize(textToDisplay, font, fontScale, fontThickness, &baseline);
-            const cv::Size finalPosition{position.x - (normalizeWidth ? textSize.width : 0), position.y + textSize.height/2};
-            cv::putText(cvMat, textToDisplay, cv::Size{finalPosition.width + shadowOffset, finalPosition.height + shadowOffset}, font, fontScale, cv::Scalar{0,0,0}, fontThickness);
+            const cv::Size finalPosition{position.x - (normalizeWidth ? textSize.width : 0),
+                                         position.y + textSize.height/2};
+            cv::putText(cvMat, textToDisplay,
+                        cv::Size{finalPosition.width + shadowOffset, finalPosition.height + shadowOffset},
+                        font, fontScale, cv::Scalar{0,0,0}, fontThickness);
            cv::putText(cvMat, textToDisplay, finalPosition, font, fontScale, color, fontThickness);
        }
        catch (const std::exception& e)
@@ -24,7 +28,8 @@ namespace op
        }
    }

-    void floatPtrToUCharCvMat(cv::Mat& uCharCvMat, const float* const floatPtrImage, const std::array<int, 3> resolutionSize)
+    void floatPtrToUCharCvMat(cv::Mat& uCharCvMat, const float* const floatPtrImage,
+                              const std::array<int, 3> resolutionSize)
    {
        try
        {
@@ -32,7 +37,8 @@ namespace op
                // float* (deep net format): C x H x W
                // cv::Mat (OpenCV format): H x W x C
            // Allocate cv::Mat if it was not initialized yet
-            if (uCharCvMat.empty() || uCharCvMat.rows != resolutionSize[1] || uCharCvMat.cols != resolutionSize[0] || uCharCvMat.type() != CV_8UC3)
+            if (uCharCvMat.empty() || uCharCvMat.rows != resolutionSize[1]
+                || uCharCvMat.cols != resolutionSize[0] || uCharCvMat.type() != CV_8UC3)
                uCharCvMat = cv::Mat(resolutionSize[1], resolutionSize[0], CV_8UC3);
            // Fill uCharCvMat from floatPtrImage
            auto* uCharPtrCvMat = (unsigned char*)(uCharCvMat.data);
@@ -47,7 +53,9 @@ namespace op
                    const auto floatPtrImageOffsetY = offsetChannelC + y*resolutionSize[0];
                    for (auto x = 0; x < resolutionSize[0]; x++)
                    {
-                        const auto value = uchar(   fastTruncate(intRound(floatPtrImage[floatPtrImageOffsetY + x]), 0, 255)   );
+                        const auto value = uchar(
+                            fastTruncate(intRound(floatPtrImage[floatPtrImageOffsetY + x]), 0, 255)
+                        );
                        uCharPtrCvMat[yOffset + x * resolutionSize[2] + c] = value;
                        // *(uCharCvMat.ptr<uchar>(y, x) + c) = value; // Slower but safer and cleaner equivalent
                    }
@@ -65,7 +73,8 @@ namespace op
        try
        {
            if (array.getNumberDimensions() != 3)
-                error("Only implemented for array.getNumberDimensions() == 3 so far.", __LINE__, __FUNCTION__, __FILE__);
+                error("Only implemented for array.getNumberDimensions() == 3 so far.",
+                      __LINE__, __FUNCTION__, __FILE__);

            if (!array.empty())
            {
@@ -90,7 +99,7 @@ namespace op
                        const auto offsetHeight = y * width;
                        for (auto x = 0 ; x < width ; x++)
                        {
-                            const auto value = uchar(   fastTruncate(intRound(arrayPtr[offsetHeight + x]), 0, 255)   );
+                            const auto value = uchar( fastTruncate(intRound(arrayPtr[offsetHeight + x]), 0, 255) );
                            cvMatROIPtr[x] = (unsigned char)(value);
                        }
                    }
@@ -124,13 +133,15 @@ namespace op
                    const auto floatPtrImageOffsetY = (floatPtrImageOffsetC + y) * width;
                    const auto originFramePtrOffsetY = y * width;
                    for (auto x = 0; x < width; x++)
-                        floatPtrImage[floatPtrImageOffsetY + x] = float(originFramePtr[(originFramePtrOffsetY + x) * channels + c]);
+                        floatPtrImage[floatPtrImageOffsetY + x] = float(originFramePtr[(originFramePtrOffsetY + x)
+                                                                        * channels + c]);
                }
            }
            // Normalizing if desired
            // floatPtrImage wrapped as cv::Mat
-                // Empirically tested - OpenCV is more efficient normalizing a whole matrix/image (it uses AVX and other optimized instruction sets)
-                // In addition, the following if statement does not copy the pointer to a cv::Mat, just wrapps it
+                // Empirically tested - OpenCV is more efficient normalizing a whole matrix/image (it uses AVX and
+                // other optimized instruction sets).
+                // In addition, the following if statement does not copy the pointer to a cv::Mat, just wrapps it.
            if (normalize)
            {
                cv::Mat floatPtrImageCvWrapper(height, width, CV_32FC3, floatPtrImage);
@@ -158,7 +169,8 @@ namespace op
        }
    }

-    cv::Mat resizeFixedAspectRatio(const cv::Mat& cvMat, const double scaleFactor, const Point<int>& targetSize, const int borderMode, const cv::Scalar& borderValue)
+    cv::Mat resizeFixedAspectRatio(const cv::Mat& cvMat, const double scaleFactor, const Point<int>& targetSize,
+                                   const int borderMode, const cv::Scalar& borderValue)
    {
        try
        {
@@ -168,7 +180,8 @@ namespace op
            M.at<double>(0,0) = scaleFactor;
            M.at<double>(1,1) = scaleFactor;
            if (scaleFactor != 1. || cvTargetSize != cvMat.size())
-                cv::warpAffine(cvMat, resultingCvMat, M, cvTargetSize, (scaleFactor < 1. ? cv::INTER_AREA : cv::INTER_CUBIC), borderMode, borderValue);
+                cv::warpAffine(cvMat, resultingCvMat, M, cvTargetSize,
+                               (scaleFactor < 1. ? cv::INTER_AREA : cv::INTER_CUBIC), borderMode, borderValue);
            else
                resultingCvMat = cvMat.clone();
            return resultingCvMat;

--- a/windows/OpenPose/OpenPose.vcxproj
+++ b/windows/OpenPose/OpenPose.vcxproj
@@ -235,6 +235,7 @@
    <ClInclude Include="..\..\include\openpose\utilities\openCv.hpp" />
    <ClInclude Include="..\..\include\openpose\utilities\pointerContainer.hpp" />
    <ClInclude Include="..\..\include\openpose\utilities\profiler.hpp" />
+    <ClInclude Include="..\..\include\openpose\utilities\standard.hpp" />
    <ClInclude Include="..\..\include\openpose\utilities\string.hpp" />
    <ClInclude Include="..\..\include\openpose\wrapper\headers.hpp" />
    <ClInclude Include="..\..\include\openpose\wrapper\wrapper.hpp" />

--- a/windows/OpenPose/OpenPose.vcxproj.filters
+++ b/windows/OpenPose/OpenPose.vcxproj.filters
@@ -497,6 +497,9 @@
    <ClInclude Include="..\..\include\openpose\utilities\profiler.hpp">
      <Filter>Header Files\utilities</Filter>
    </ClInclude>
+    <ClInclude Include="..\..\include\openpose\utilities\standard.hpp">
+      <Filter>Header Files\utilities</Filter>
+    </ClInclude>
    <ClInclude Include="..\..\include\openpose\utilities\string.hpp">
      <Filter>Header Files\utilities</Filter>
    </ClInclude>

--- a/windows/OpenPose3DReconstruction/OpenPose3DReconstruction.vcxproj
+++ b/windows/OpenPose3DReconstruction/OpenPose3DReconstruction.vcxproj
@@ -83,7 +83,7 @@
      <FunctionLevelLinking>true</FunctionLevelLinking>
      <IntrinsicFunctions>true</IntrinsicFunctions>
      <SDLCheck>true</SDLCheck>
-      <AdditionalIncludeDirectories>..\..\include;..\..\3rdparty\windows\opencv\include;..\..\3rdparty\windows\caffe\include;..\..\3rdparty\windows\caffe\include2;..\..\3rdparty\windows\caffe3rdparty\include;..\..\3rdparty\windows\caffe3rdparty\include\boost-1_61;$(CUDA_PATH_V8_0)\include;..\..\examples_beta\openpose3d\include;..\..\3rdparty\windows\freeglut\include;..\..\3rdparty\windows\spinnaker\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+      <AdditionalIncludeDirectories>..\..\include;..\..\3rdparty\windows\opencv\include;..\..\3rdparty\windows\caffe3rdparty\include;..\..\examples_beta\openpose3d\include;..\..\3rdparty\windows\freeglut\include;..\..\3rdparty\windows\spinnaker\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
      <PreprocessorDefinitions>_CRT_SECURE_NO_WARNINGS;CMAKE_WINDOWS_BUILD;GLOG_NO_ABBREVIATED_SEVERITIES;GOOGLE_GLOG_DLL_DECL=__declspec(dllimport);GOOGLE_GLOG_DLL_DECL_FOR_UNITTESTS=__declspec(dllimport);CMAKE_INTDIR="Release";NDEBUG</PreprocessorDefinitions>
      <ObjectFileName>$(IntDir)\obj\relDir\relDir\%(RelativeDir)\%(Filename)%(Extension).obj</ObjectFileName>
      <MultiProcessorCompilation>true</MultiProcessorCompilation>
@@ -94,7 +94,7 @@
      <EnableCOMDATFolding>true</EnableCOMDATFolding>
      <OptimizeReferences>true</OptimizeReferences>
      <AdditionalLibraryDirectories>$(CUDA_PATH_V8_0)\lib\x64\;</AdditionalLibraryDirectories>
-      <AdditionalDependencies>..\..\3rdparty\windows\caffe\lib\caffe.lib;..\..\3rdparty\windows\caffe\lib\caffeproto.lib;..\..\3rdparty\windows\caffe3rdparty\lib\boost_filesystem-vc140-mt-1_61.lib;..\..\3rdparty\windows\caffe3rdparty\lib\boost_system-vc140-mt-1_61.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffehdf5.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffehdf5_hl.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffezlib.lib;..\..\3rdparty\windows\caffe3rdparty\lib\gflags.lib;..\..\3rdparty\windows\caffe3rdparty\lib\glog.lib;..\..\3rdparty\windows\caffe3rdparty\lib\leveldb.lib;..\..\3rdparty\windows\caffe3rdparty\lib\libopenblas.dll.a;..\..\3rdparty\windows\caffe3rdparty\lib\libprotobuf.lib;..\..\3rdparty\windows\caffe3rdparty\lib\lmdb.lib;..\..\3rdparty\windows\caffe3rdparty\lib\snappy.lib;..\..\3rdparty\windows\opencv\x64\vc14\lib\opencv_world310.lib;cublas.lib;cublas_device.lib;cudart.lib;cudnn.lib;curand.lib;ntdll.lib;shlwapi.lib;..\x64\Release\OpenPose.lib;..\..\3rdparty\windows\freeglut\lib\freeglut.lib;..\..\3rdparty\windows\spinnaker\lib\Spinnaker_$(PlatformToolset).lib;%(AdditionalDependencies)</AdditionalDependencies>
+      <AdditionalDependencies>..\..\3rdparty\windows\caffe3rdparty\lib\gflags.lib;..\..\3rdparty\windows\opencv\x64\vc14\lib\opencv_world310.lib;..\x64\Release\OpenPose.lib;..\..\3rdparty\windows\freeglut\lib\freeglut.lib;..\..\3rdparty\windows\spinnaker\lib\Spinnaker_$(PlatformToolset).lib;%(AdditionalDependencies)</AdditionalDependencies>
    </Link>
  </ItemDefinitionGroup>
  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />

--- a/windows/OpenPoseDemo/OpenPoseDemo.vcxproj
+++ b/windows/OpenPoseDemo/OpenPoseDemo.vcxproj
@@ -69,7 +69,7 @@
      <FunctionLevelLinking>true</FunctionLevelLinking>
      <IntrinsicFunctions>true</IntrinsicFunctions>
      <SDLCheck>true</SDLCheck>
-      <AdditionalIncludeDirectories>..\..\include;..\..\3rdparty\windows\opencv\include;..\..\3rdparty\windows\caffe\include;..\..\3rdparty\windows\caffe\include2;..\..\3rdparty\windows\caffe3rdparty\include;..\..\3rdparty\windows\caffe3rdparty\include\boost-1_61;$(CUDA_PATH_V8_0)\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+      <AdditionalIncludeDirectories>..\..\include;..\..\3rdparty\windows\opencv\include;..\..\3rdparty\windows\caffe3rdparty\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
      <PreprocessorDefinitions>_CRT_SECURE_NO_WARNINGS;CMAKE_WINDOWS_BUILD;GLOG_NO_ABBREVIATED_SEVERITIES;GOOGLE_GLOG_DLL_DECL=__declspec(dllimport);GOOGLE_GLOG_DLL_DECL_FOR_UNITTESTS=__declspec(dllimport);CMAKE_INTDIR="Release";NDEBUG</PreprocessorDefinitions>
      <ObjectFileName>$(IntDir)\obj\relDir\relDir\%(RelativeDir)\%(Filename)%(Extension).obj</ObjectFileName>
      <MultiProcessorCompilation>true</MultiProcessorCompilation>
@@ -80,7 +80,7 @@
      <EnableCOMDATFolding>true</EnableCOMDATFolding>
      <OptimizeReferences>true</OptimizeReferences>
      <AdditionalLibraryDirectories>$(CUDA_PATH_V8_0)\lib\x64\;</AdditionalLibraryDirectories>
-      <AdditionalDependencies>..\..\3rdparty\windows\caffe\lib\caffe.lib;..\..\3rdparty\windows\caffe\lib\caffeproto.lib;..\..\3rdparty\windows\caffe3rdparty\lib\boost_filesystem-vc140-mt-1_61.lib;..\..\3rdparty\windows\caffe3rdparty\lib\boost_system-vc140-mt-1_61.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffehdf5.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffehdf5_hl.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffezlib.lib;..\..\3rdparty\windows\caffe3rdparty\lib\gflags.lib;..\..\3rdparty\windows\caffe3rdparty\lib\glog.lib;..\..\3rdparty\windows\caffe3rdparty\lib\leveldb.lib;..\..\3rdparty\windows\caffe3rdparty\lib\libopenblas.dll.a;..\..\3rdparty\windows\caffe3rdparty\lib\libprotobuf.lib;..\..\3rdparty\windows\caffe3rdparty\lib\lmdb.lib;..\..\3rdparty\windows\caffe3rdparty\lib\snappy.lib;..\..\3rdparty\windows\opencv\x64\vc14\lib\opencv_world310.lib;cublas.lib;cublas_device.lib;cudart.lib;cudnn.lib;curand.lib;ntdll.lib;shlwapi.lib;..\x64\Release\OpenPose.lib;%(AdditionalDependencies)</AdditionalDependencies>
+      <AdditionalDependencies>..\..\3rdparty\windows\caffe3rdparty\lib\gflags.lib;..\..\3rdparty\windows\opencv\x64\vc14\lib\opencv_world310.lib;..\x64\Release\OpenPose.lib;%(AdditionalDependencies)</AdditionalDependencies>
    </Link>
  </ItemDefinitionGroup>
  <ItemGroup>

--- a/windows/TutorialAddModule/1_custom_post_processing.vcxproj
+++ b/windows/TutorialAddModule/1_custom_post_processing.vcxproj
@@ -77,7 +77,7 @@
      <FunctionLevelLinking>true</FunctionLevelLinking>
      <IntrinsicFunctions>true</IntrinsicFunctions>
      <SDLCheck>true</SDLCheck>
-      <AdditionalIncludeDirectories>..\..\include;..\..\3rdparty\windows\opencv\include;..\..\3rdparty\windows\caffe\include;..\..\3rdparty\windows\caffe\include2;..\..\3rdparty\windows\caffe3rdparty\include;..\..\3rdparty\windows\caffe3rdparty\include\boost-1_61;$(CUDA_PATH_V8_0)\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+      <AdditionalIncludeDirectories>..\..\include;..\..\3rdparty\windows\opencv\include;..\..\3rdparty\windows\caffe3rdparty\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
      <PreprocessorDefinitions>_CRT_SECURE_NO_WARNINGS;CMAKE_WINDOWS_BUILD;GLOG_NO_ABBREVIATED_SEVERITIES;GOOGLE_GLOG_DLL_DECL=__declspec(dllimport);GOOGLE_GLOG_DLL_DECL_FOR_UNITTESTS=__declspec(dllimport);CMAKE_INTDIR="Release";NDEBUG</PreprocessorDefinitions>
      <ObjectFileName>$(IntDir)\obj\relDir\relDir\%(RelativeDir)\%(Filename)%(Extension).obj</ObjectFileName>
      <MultiProcessorCompilation>true</MultiProcessorCompilation>
@@ -88,7 +88,7 @@
      <EnableCOMDATFolding>true</EnableCOMDATFolding>
      <OptimizeReferences>true</OptimizeReferences>
      <AdditionalLibraryDirectories>$(CUDA_PATH_V8_0)\lib\x64\;</AdditionalLibraryDirectories>
-      <AdditionalDependencies>..\..\3rdparty\windows\caffe\lib\caffe.lib;..\..\3rdparty\windows\caffe\lib\caffeproto.lib;..\..\3rdparty\windows\caffe3rdparty\lib\boost_filesystem-vc140-mt-1_61.lib;..\..\3rdparty\windows\caffe3rdparty\lib\boost_system-vc140-mt-1_61.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffehdf5.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffehdf5_hl.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffezlib.lib;..\..\3rdparty\windows\caffe3rdparty\lib\gflags.lib;..\..\3rdparty\windows\caffe3rdparty\lib\glog.lib;..\..\3rdparty\windows\caffe3rdparty\lib\leveldb.lib;..\..\3rdparty\windows\caffe3rdparty\lib\libopenblas.dll.a;..\..\3rdparty\windows\caffe3rdparty\lib\libprotobuf.lib;..\..\3rdparty\windows\caffe3rdparty\lib\lmdb.lib;..\..\3rdparty\windows\caffe3rdparty\lib\snappy.lib;..\..\3rdparty\windows\opencv\x64\vc14\lib\opencv_world310.lib;cublas.lib;cublas_device.lib;cudart.lib;cudnn.lib;curand.lib;ntdll.lib;shlwapi.lib;..\x64\Release\OpenPose.lib;%(AdditionalDependencies)</AdditionalDependencies>
+      <AdditionalDependencies>..\..\3rdparty\windows\caffe3rdparty\lib\gflags.lib;..\..\3rdparty\windows\opencv\x64\vc14\lib\opencv_world310.lib;..\x64\Release\OpenPose.lib;%(AdditionalDependencies)</AdditionalDependencies>
    </Link>
  </ItemDefinitionGroup>
  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />

--- a/windows/TutorialPose/1_extract_from_image.vcxproj
+++ b/windows/TutorialPose/1_extract_from_image.vcxproj
@@ -72,7 +72,7 @@
      <FunctionLevelLinking>true</FunctionLevelLinking>
      <IntrinsicFunctions>true</IntrinsicFunctions>
      <SDLCheck>true</SDLCheck>
-      <AdditionalIncludeDirectories>..\..\include;..\..\3rdparty\windows\opencv\include;..\..\3rdparty\windows\caffe\include;..\..\3rdparty\windows\caffe\include2;..\..\3rdparty\windows\caffe3rdparty\include;..\..\3rdparty\windows\caffe3rdparty\include\boost-1_61;$(CUDA_PATH_V8_0)\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+      <AdditionalIncludeDirectories>..\..\include;..\..\3rdparty\windows\opencv\include;..\..\3rdparty\windows\caffe3rdparty\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
      <PreprocessorDefinitions>_CRT_SECURE_NO_WARNINGS;CMAKE_WINDOWS_BUILD;GLOG_NO_ABBREVIATED_SEVERITIES;GOOGLE_GLOG_DLL_DECL=__declspec(dllimport);GOOGLE_GLOG_DLL_DECL_FOR_UNITTESTS=__declspec(dllimport);CMAKE_INTDIR="Release";NDEBUG</PreprocessorDefinitions>
      <ObjectFileName>$(IntDir)\obj\relDir\relDir\%(RelativeDir)\%(Filename)%(Extension).obj</ObjectFileName>
      <MultiProcessorCompilation>true</MultiProcessorCompilation>
@@ -83,7 +83,7 @@
      <EnableCOMDATFolding>true</EnableCOMDATFolding>
      <OptimizeReferences>true</OptimizeReferences>
      <AdditionalLibraryDirectories>$(CUDA_PATH_V8_0)\lib\x64\;</AdditionalLibraryDirectories>
-      <AdditionalDependencies>..\..\3rdparty\windows\caffe\lib\caffe.lib;..\..\3rdparty\windows\caffe\lib\caffeproto.lib;..\..\3rdparty\windows\caffe3rdparty\lib\boost_filesystem-vc140-mt-1_61.lib;..\..\3rdparty\windows\caffe3rdparty\lib\boost_system-vc140-mt-1_61.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffehdf5.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffehdf5_hl.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffezlib.lib;..\..\3rdparty\windows\caffe3rdparty\lib\gflags.lib;..\..\3rdparty\windows\caffe3rdparty\lib\glog.lib;..\..\3rdparty\windows\caffe3rdparty\lib\leveldb.lib;..\..\3rdparty\windows\caffe3rdparty\lib\libopenblas.dll.a;..\..\3rdparty\windows\caffe3rdparty\lib\libprotobuf.lib;..\..\3rdparty\windows\caffe3rdparty\lib\lmdb.lib;..\..\3rdparty\windows\caffe3rdparty\lib\snappy.lib;..\..\3rdparty\windows\opencv\x64\vc14\lib\opencv_world310.lib;cublas.lib;cublas_device.lib;cudart.lib;cudnn.lib;curand.lib;ntdll.lib;shlwapi.lib;..\x64\Release\OpenPose.lib;%(AdditionalDependencies)</AdditionalDependencies>
+      <AdditionalDependencies>..\..\3rdparty\windows\caffe3rdparty\lib\gflags.lib;..\..\3rdparty\windows\opencv\x64\vc14\lib\opencv_world310.lib;..\x64\Release\OpenPose.lib;%(AdditionalDependencies)</AdditionalDependencies>
    </Link>
  </ItemDefinitionGroup>
  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />

--- a/windows/TutorialPose/2_extract_pose_or_heatmat_from_image.vcxproj
+++ b/windows/TutorialPose/2_extract_pose_or_heatmat_from_image.vcxproj
@@ -72,7 +72,7 @@
      <FunctionLevelLinking>true</FunctionLevelLinking>
      <IntrinsicFunctions>true</IntrinsicFunctions>
      <SDLCheck>true</SDLCheck>
-      <AdditionalIncludeDirectories>..\..\include;..\..\3rdparty\windows\opencv\include;..\..\3rdparty\windows\caffe\include;..\..\3rdparty\windows\caffe\include2;..\..\3rdparty\windows\caffe3rdparty\include;..\..\3rdparty\windows\caffe3rdparty\include\boost-1_61;$(CUDA_PATH_V8_0)\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+      <AdditionalIncludeDirectories>..\..\include;..\..\3rdparty\windows\opencv\include;..\..\3rdparty\windows\caffe3rdparty\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
      <PreprocessorDefinitions>_CRT_SECURE_NO_WARNINGS;CMAKE_WINDOWS_BUILD;GLOG_NO_ABBREVIATED_SEVERITIES;GOOGLE_GLOG_DLL_DECL=__declspec(dllimport);GOOGLE_GLOG_DLL_DECL_FOR_UNITTESTS=__declspec(dllimport);CMAKE_INTDIR="Release";NDEBUG</PreprocessorDefinitions>
      <ObjectFileName>$(IntDir)\obj\relDir\relDir\%(RelativeDir)\%(Filename)%(Extension).obj</ObjectFileName>
      <MultiProcessorCompilation>true</MultiProcessorCompilation>
@@ -83,7 +83,7 @@
      <EnableCOMDATFolding>true</EnableCOMDATFolding>
      <OptimizeReferences>true</OptimizeReferences>
      <AdditionalLibraryDirectories>$(CUDA_PATH_V8_0)\lib\x64\;</AdditionalLibraryDirectories>
-      <AdditionalDependencies>..\..\3rdparty\windows\caffe\lib\caffe.lib;..\..\3rdparty\windows\caffe\lib\caffeproto.lib;..\..\3rdparty\windows\caffe3rdparty\lib\boost_filesystem-vc140-mt-1_61.lib;..\..\3rdparty\windows\caffe3rdparty\lib\boost_system-vc140-mt-1_61.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffehdf5.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffehdf5_hl.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffezlib.lib;..\..\3rdparty\windows\caffe3rdparty\lib\gflags.lib;..\..\3rdparty\windows\caffe3rdparty\lib\glog.lib;..\..\3rdparty\windows\caffe3rdparty\lib\leveldb.lib;..\..\3rdparty\windows\caffe3rdparty\lib\libopenblas.dll.a;..\..\3rdparty\windows\caffe3rdparty\lib\libprotobuf.lib;..\..\3rdparty\windows\caffe3rdparty\lib\lmdb.lib;..\..\3rdparty\windows\caffe3rdparty\lib\snappy.lib;..\..\3rdparty\windows\opencv\x64\vc14\lib\opencv_world310.lib;cublas.lib;cublas_device.lib;cudart.lib;cudnn.lib;curand.lib;ntdll.lib;shlwapi.lib;..\x64\Release\OpenPose.lib;%(AdditionalDependencies)</AdditionalDependencies>
+      <AdditionalDependencies>..\..\3rdparty\windows\caffe3rdparty\lib\gflags.lib;..\..\3rdparty\windows\opencv\x64\vc14\lib\opencv_world310.lib;..\x64\Release\OpenPose.lib;%(AdditionalDependencies)</AdditionalDependencies>
    </Link>
  </ItemDefinitionGroup>
  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />

--- a/windows/TutorialThread/1_openpose_read_and_display.vcxproj
+++ b/windows/TutorialThread/1_openpose_read_and_display.vcxproj
@@ -72,7 +72,7 @@
      <FunctionLevelLinking>true</FunctionLevelLinking>
      <IntrinsicFunctions>true</IntrinsicFunctions>
      <SDLCheck>true</SDLCheck>
-      <AdditionalIncludeDirectories>..\..\include;..\..\3rdparty\windows\opencv\include;..\..\3rdparty\windows\caffe\include;..\..\3rdparty\windows\caffe\include2;..\..\3rdparty\windows\caffe3rdparty\include;..\..\3rdparty\windows\caffe3rdparty\include\boost-1_61;$(CUDA_PATH_V8_0)\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+      <AdditionalIncludeDirectories>..\..\include;..\..\3rdparty\windows\opencv\include;..\..\3rdparty\windows\caffe3rdparty\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
      <PreprocessorDefinitions>_CRT_SECURE_NO_WARNINGS;CMAKE_WINDOWS_BUILD;GLOG_NO_ABBREVIATED_SEVERITIES;GOOGLE_GLOG_DLL_DECL=__declspec(dllimport);GOOGLE_GLOG_DLL_DECL_FOR_UNITTESTS=__declspec(dllimport);CMAKE_INTDIR="Release";NDEBUG</PreprocessorDefinitions>
      <ObjectFileName>$(IntDir)\obj\relDir\relDir\%(RelativeDir)\%(Filename)%(Extension).obj</ObjectFileName>
      <MultiProcessorCompilation>true</MultiProcessorCompilation>
@@ -83,7 +83,7 @@
      <EnableCOMDATFolding>true</EnableCOMDATFolding>
      <OptimizeReferences>true</OptimizeReferences>
      <AdditionalLibraryDirectories>$(CUDA_PATH_V8_0)\lib\x64\;</AdditionalLibraryDirectories>
-      <AdditionalDependencies>..\..\3rdparty\windows\caffe\lib\caffe.lib;..\..\3rdparty\windows\caffe\lib\caffeproto.lib;..\..\3rdparty\windows\caffe3rdparty\lib\boost_filesystem-vc140-mt-1_61.lib;..\..\3rdparty\windows\caffe3rdparty\lib\boost_system-vc140-mt-1_61.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffehdf5.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffehdf5_hl.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffezlib.lib;..\..\3rdparty\windows\caffe3rdparty\lib\gflags.lib;..\..\3rdparty\windows\caffe3rdparty\lib\glog.lib;..\..\3rdparty\windows\caffe3rdparty\lib\leveldb.lib;..\..\3rdparty\windows\caffe3rdparty\lib\libopenblas.dll.a;..\..\3rdparty\windows\caffe3rdparty\lib\libprotobuf.lib;..\..\3rdparty\windows\caffe3rdparty\lib\lmdb.lib;..\..\3rdparty\windows\caffe3rdparty\lib\snappy.lib;..\..\3rdparty\windows\opencv\x64\vc14\lib\opencv_world310.lib;cublas.lib;cublas_device.lib;cudart.lib;cudnn.lib;curand.lib;ntdll.lib;shlwapi.lib;..\x64\Release\OpenPose.lib;%(AdditionalDependencies)</AdditionalDependencies>
+      <AdditionalDependencies>..\..\3rdparty\windows\caffe3rdparty\lib\gflags.lib;..\..\3rdparty\windows\opencv\x64\vc14\lib\opencv_world310.lib;..\x64\Release\OpenPose.lib;%(AdditionalDependencies)</AdditionalDependencies>
    </Link>
  </ItemDefinitionGroup>
  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />

--- a/windows/TutorialThread/2_user_processing_function.vcxproj
+++ b/windows/TutorialThread/2_user_processing_function.vcxproj
@@ -72,7 +72,7 @@
      <FunctionLevelLinking>true</FunctionLevelLinking>
      <IntrinsicFunctions>true</IntrinsicFunctions>
      <SDLCheck>true</SDLCheck>
-      <AdditionalIncludeDirectories>..\..\include;..\..\3rdparty\windows\opencv\include;..\..\3rdparty\windows\caffe\include;..\..\3rdparty\windows\caffe\include2;..\..\3rdparty\windows\caffe3rdparty\include;..\..\3rdparty\windows\caffe3rdparty\include\boost-1_61;$(CUDA_PATH_V8_0)\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+      <AdditionalIncludeDirectories>..\..\include;..\..\3rdparty\windows\opencv\include;..\..\3rdparty\windows\caffe3rdparty\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
      <PreprocessorDefinitions>_CRT_SECURE_NO_WARNINGS;CMAKE_WINDOWS_BUILD;GLOG_NO_ABBREVIATED_SEVERITIES;GOOGLE_GLOG_DLL_DECL=__declspec(dllimport);GOOGLE_GLOG_DLL_DECL_FOR_UNITTESTS=__declspec(dllimport);CMAKE_INTDIR="Release";NDEBUG</PreprocessorDefinitions>
      <ObjectFileName>$(IntDir)\obj\relDir\relDir\%(RelativeDir)\%(Filename)%(Extension).obj</ObjectFileName>
      <MultiProcessorCompilation>true</MultiProcessorCompilation>
@@ -83,7 +83,7 @@
      <EnableCOMDATFolding>true</EnableCOMDATFolding>
      <OptimizeReferences>true</OptimizeReferences>
      <AdditionalLibraryDirectories>$(CUDA_PATH_V8_0)\lib\x64\;</AdditionalLibraryDirectories>
-      <AdditionalDependencies>..\..\3rdparty\windows\caffe\lib\caffe.lib;..\..\3rdparty\windows\caffe\lib\caffeproto.lib;..\..\3rdparty\windows\caffe3rdparty\lib\boost_filesystem-vc140-mt-1_61.lib;..\..\3rdparty\windows\caffe3rdparty\lib\boost_system-vc140-mt-1_61.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffehdf5.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffehdf5_hl.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffezlib.lib;..\..\3rdparty\windows\caffe3rdparty\lib\gflags.lib;..\..\3rdparty\windows\caffe3rdparty\lib\glog.lib;..\..\3rdparty\windows\caffe3rdparty\lib\leveldb.lib;..\..\3rdparty\windows\caffe3rdparty\lib\libopenblas.dll.a;..\..\3rdparty\windows\caffe3rdparty\lib\libprotobuf.lib;..\..\3rdparty\windows\caffe3rdparty\lib\lmdb.lib;..\..\3rdparty\windows\caffe3rdparty\lib\snappy.lib;..\..\3rdparty\windows\opencv\x64\vc14\lib\opencv_world310.lib;cublas.lib;cublas_device.lib;cudart.lib;cudnn.lib;curand.lib;ntdll.lib;shlwapi.lib;..\x64\Release\OpenPose.lib;%(AdditionalDependencies)</AdditionalDependencies>
+      <AdditionalDependencies>..\..\3rdparty\windows\caffe3rdparty\lib\gflags.lib;..\..\3rdparty\windows\opencv\x64\vc14\lib\opencv_world310.lib;..\x64\Release\OpenPose.lib;%(AdditionalDependencies)</AdditionalDependencies>
    </Link>
  </ItemDefinitionGroup>
  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />

--- a/windows/TutorialThread/3_user_input_processing_and_output.vcxproj
+++ b/windows/TutorialThread/3_user_input_processing_and_output.vcxproj
@@ -72,7 +72,7 @@
      <FunctionLevelLinking>true</FunctionLevelLinking>
      <IntrinsicFunctions>true</IntrinsicFunctions>
      <SDLCheck>true</SDLCheck>
-      <AdditionalIncludeDirectories>..\..\include;..\..\3rdparty\windows\opencv\include;..\..\3rdparty\windows\caffe\include;..\..\3rdparty\windows\caffe\include2;..\..\3rdparty\windows\caffe3rdparty\include;..\..\3rdparty\windows\caffe3rdparty\include\boost-1_61;$(CUDA_PATH_V8_0)\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+      <AdditionalIncludeDirectories>..\..\include;..\..\3rdparty\windows\opencv\include;..\..\3rdparty\windows\caffe3rdparty\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
      <PreprocessorDefinitions>_CRT_SECURE_NO_WARNINGS;CMAKE_WINDOWS_BUILD;GLOG_NO_ABBREVIATED_SEVERITIES;GOOGLE_GLOG_DLL_DECL=__declspec(dllimport);GOOGLE_GLOG_DLL_DECL_FOR_UNITTESTS=__declspec(dllimport);CMAKE_INTDIR="Release";NDEBUG</PreprocessorDefinitions>
      <ObjectFileName>$(IntDir)\obj\relDir\relDir\%(RelativeDir)\%(Filename)%(Extension).obj</ObjectFileName>
      <MultiProcessorCompilation>true</MultiProcessorCompilation>
@@ -83,7 +83,7 @@
      <EnableCOMDATFolding>true</EnableCOMDATFolding>
      <OptimizeReferences>true</OptimizeReferences>
      <AdditionalLibraryDirectories>$(CUDA_PATH_V8_0)\lib\x64\;</AdditionalLibraryDirectories>
-      <AdditionalDependencies>..\..\3rdparty\windows\caffe\lib\caffe.lib;..\..\3rdparty\windows\caffe\lib\caffeproto.lib;..\..\3rdparty\windows\caffe3rdparty\lib\boost_filesystem-vc140-mt-1_61.lib;..\..\3rdparty\windows\caffe3rdparty\lib\boost_system-vc140-mt-1_61.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffehdf5.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffehdf5_hl.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffezlib.lib;..\..\3rdparty\windows\caffe3rdparty\lib\gflags.lib;..\..\3rdparty\windows\caffe3rdparty\lib\glog.lib;..\..\3rdparty\windows\caffe3rdparty\lib\leveldb.lib;..\..\3rdparty\windows\caffe3rdparty\lib\libopenblas.dll.a;..\..\3rdparty\windows\caffe3rdparty\lib\libprotobuf.lib;..\..\3rdparty\windows\caffe3rdparty\lib\lmdb.lib;..\..\3rdparty\windows\caffe3rdparty\lib\snappy.lib;..\..\3rdparty\windows\opencv\x64\vc14\lib\opencv_world310.lib;cublas.lib;cublas_device.lib;cudart.lib;cudnn.lib;curand.lib;ntdll.lib;shlwapi.lib;..\x64\Release\OpenPose.lib;%(AdditionalDependencies)</AdditionalDependencies>
+      <AdditionalDependencies>..\..\3rdparty\windows\caffe3rdparty\lib\gflags.lib;..\..\3rdparty\windows\opencv\x64\vc14\lib\opencv_world310.lib;..\x64\Release\OpenPose.lib;%(AdditionalDependencies)</AdditionalDependencies>
    </Link>
  </ItemDefinitionGroup>
  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />

--- a/windows/TutorialThread/4_user_input_processing_output_and_datum.vcxproj
+++ b/windows/TutorialThread/4_user_input_processing_output_and_datum.vcxproj
@@ -72,7 +72,7 @@
      <FunctionLevelLinking>true</FunctionLevelLinking>
      <IntrinsicFunctions>true</IntrinsicFunctions>
      <SDLCheck>true</SDLCheck>
-      <AdditionalIncludeDirectories>..\..\include;..\..\3rdparty\windows\opencv\include;..\..\3rdparty\windows\caffe\include;..\..\3rdparty\windows\caffe\include2;..\..\3rdparty\windows\caffe3rdparty\include;..\..\3rdparty\windows\caffe3rdparty\include\boost-1_61;$(CUDA_PATH_V8_0)\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+      <AdditionalIncludeDirectories>..\..\include;..\..\3rdparty\windows\opencv\include;..\..\3rdparty\windows\caffe3rdparty\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
      <PreprocessorDefinitions>_CRT_SECURE_NO_WARNINGS;CMAKE_WINDOWS_BUILD;GLOG_NO_ABBREVIATED_SEVERITIES;GOOGLE_GLOG_DLL_DECL=__declspec(dllimport);GOOGLE_GLOG_DLL_DECL_FOR_UNITTESTS=__declspec(dllimport);CMAKE_INTDIR="Release";NDEBUG</PreprocessorDefinitions>
      <ObjectFileName>$(IntDir)\obj\relDir\relDir\%(RelativeDir)\%(Filename)%(Extension).obj</ObjectFileName>
      <MultiProcessorCompilation>true</MultiProcessorCompilation>
@@ -83,7 +83,7 @@
      <EnableCOMDATFolding>true</EnableCOMDATFolding>
      <OptimizeReferences>true</OptimizeReferences>
      <AdditionalLibraryDirectories>$(CUDA_PATH_V8_0)\lib\x64\;</AdditionalLibraryDirectories>
-      <AdditionalDependencies>..\..\3rdparty\windows\caffe\lib\caffe.lib;..\..\3rdparty\windows\caffe\lib\caffeproto.lib;..\..\3rdparty\windows\caffe3rdparty\lib\boost_filesystem-vc140-mt-1_61.lib;..\..\3rdparty\windows\caffe3rdparty\lib\boost_system-vc140-mt-1_61.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffehdf5.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffehdf5_hl.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffezlib.lib;..\..\3rdparty\windows\caffe3rdparty\lib\gflags.lib;..\..\3rdparty\windows\caffe3rdparty\lib\glog.lib;..\..\3rdparty\windows\caffe3rdparty\lib\leveldb.lib;..\..\3rdparty\windows\caffe3rdparty\lib\libopenblas.dll.a;..\..\3rdparty\windows\caffe3rdparty\lib\libprotobuf.lib;..\..\3rdparty\windows\caffe3rdparty\lib\lmdb.lib;..\..\3rdparty\windows\caffe3rdparty\lib\snappy.lib;..\..\3rdparty\windows\opencv\x64\vc14\lib\opencv_world310.lib;cublas.lib;cublas_device.lib;cudart.lib;cudnn.lib;curand.lib;ntdll.lib;shlwapi.lib;..\x64\Release\OpenPose.lib;%(AdditionalDependencies)</AdditionalDependencies>
+      <AdditionalDependencies>..\..\3rdparty\windows\caffe3rdparty\lib\gflags.lib;..\..\3rdparty\windows\opencv\x64\vc14\lib\opencv_world310.lib;..\x64\Release\OpenPose.lib;%(AdditionalDependencies)</AdditionalDependencies>
    </Link>
  </ItemDefinitionGroup>
  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />

--- a/windows/TutorialWrapper/1_user_asynchronous_output.vcxproj
+++ b/windows/TutorialWrapper/1_user_asynchronous_output.vcxproj
@@ -72,7 +72,7 @@
      <FunctionLevelLinking>true</FunctionLevelLinking>
      <IntrinsicFunctions>true</IntrinsicFunctions>
      <SDLCheck>true</SDLCheck>
-      <AdditionalIncludeDirectories>..\..\include;..\..\3rdparty\windows\opencv\include;..\..\3rdparty\windows\caffe\include;..\..\3rdparty\windows\caffe\include2;..\..\3rdparty\windows\caffe3rdparty\include;..\..\3rdparty\windows\caffe3rdparty\include\boost-1_61;$(CUDA_PATH_V8_0)\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+      <AdditionalIncludeDirectories>..\..\include;..\..\3rdparty\windows\opencv\include;..\..\3rdparty\windows\caffe3rdparty\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
      <PreprocessorDefinitions>_CRT_SECURE_NO_WARNINGS;CMAKE_WINDOWS_BUILD;GLOG_NO_ABBREVIATED_SEVERITIES;GOOGLE_GLOG_DLL_DECL=__declspec(dllimport);GOOGLE_GLOG_DLL_DECL_FOR_UNITTESTS=__declspec(dllimport);CMAKE_INTDIR="Release";NDEBUG</PreprocessorDefinitions>
      <ObjectFileName>$(IntDir)\obj\relDir\relDir\%(RelativeDir)\%(Filename)%(Extension).obj</ObjectFileName>
      <MultiProcessorCompilation>true</MultiProcessorCompilation>
@@ -83,7 +83,7 @@
      <EnableCOMDATFolding>true</EnableCOMDATFolding>
      <OptimizeReferences>true</OptimizeReferences>
      <AdditionalLibraryDirectories>$(CUDA_PATH_V8_0)\lib\x64\;</AdditionalLibraryDirectories>
-      <AdditionalDependencies>..\..\3rdparty\windows\caffe\lib\caffe.lib;..\..\3rdparty\windows\caffe\lib\caffeproto.lib;..\..\3rdparty\windows\caffe3rdparty\lib\boost_filesystem-vc140-mt-1_61.lib;..\..\3rdparty\windows\caffe3rdparty\lib\boost_system-vc140-mt-1_61.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffehdf5.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffehdf5_hl.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffezlib.lib;..\..\3rdparty\windows\caffe3rdparty\lib\gflags.lib;..\..\3rdparty\windows\caffe3rdparty\lib\glog.lib;..\..\3rdparty\windows\caffe3rdparty\lib\leveldb.lib;..\..\3rdparty\windows\caffe3rdparty\lib\libopenblas.dll.a;..\..\3rdparty\windows\caffe3rdparty\lib\libprotobuf.lib;..\..\3rdparty\windows\caffe3rdparty\lib\lmdb.lib;..\..\3rdparty\windows\caffe3rdparty\lib\snappy.lib;..\..\3rdparty\windows\opencv\x64\vc14\lib\opencv_world310.lib;cublas.lib;cublas_device.lib;cudart.lib;cudnn.lib;curand.lib;ntdll.lib;shlwapi.lib;..\x64\Release\OpenPose.lib;%(AdditionalDependencies)</AdditionalDependencies>
+      <AdditionalDependencies>..\..\3rdparty\windows\caffe3rdparty\lib\gflags.lib;..\..\3rdparty\windows\opencv\x64\vc14\lib\opencv_world310.lib;..\x64\Release\OpenPose.lib;%(AdditionalDependencies)</AdditionalDependencies>
    </Link>
  </ItemDefinitionGroup>
  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />

--- a/windows/TutorialWrapper/2_user_synchronous.vcxproj
+++ b/windows/TutorialWrapper/2_user_synchronous.vcxproj
@@ -72,7 +72,7 @@
      <FunctionLevelLinking>true</FunctionLevelLinking>
      <IntrinsicFunctions>true</IntrinsicFunctions>
      <SDLCheck>true</SDLCheck>
-      <AdditionalIncludeDirectories>..\..\include;..\..\3rdparty\windows\opencv\include;..\..\3rdparty\windows\caffe\include;..\..\3rdparty\windows\caffe\include2;..\..\3rdparty\windows\caffe3rdparty\include;..\..\3rdparty\windows\caffe3rdparty\include\boost-1_61;$(CUDA_PATH_V8_0)\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+      <AdditionalIncludeDirectories>..\..\include;..\..\3rdparty\windows\opencv\include;..\..\3rdparty\windows\caffe3rdparty\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
      <PreprocessorDefinitions>_CRT_SECURE_NO_WARNINGS;CMAKE_WINDOWS_BUILD;GLOG_NO_ABBREVIATED_SEVERITIES;GOOGLE_GLOG_DLL_DECL=__declspec(dllimport);GOOGLE_GLOG_DLL_DECL_FOR_UNITTESTS=__declspec(dllimport);CMAKE_INTDIR="Release";NDEBUG</PreprocessorDefinitions>
      <ObjectFileName>$(IntDir)\obj\relDir\relDir\%(RelativeDir)\%(Filename)%(Extension).obj</ObjectFileName>
      <MultiProcessorCompilation>true</MultiProcessorCompilation>
@@ -83,7 +83,7 @@
      <EnableCOMDATFolding>true</EnableCOMDATFolding>
      <OptimizeReferences>true</OptimizeReferences>
      <AdditionalLibraryDirectories>$(CUDA_PATH_V8_0)\lib\x64\;</AdditionalLibraryDirectories>
-      <AdditionalDependencies>..\..\3rdparty\windows\caffe\lib\caffe.lib;..\..\3rdparty\windows\caffe\lib\caffeproto.lib;..\..\3rdparty\windows\caffe3rdparty\lib\boost_filesystem-vc140-mt-1_61.lib;..\..\3rdparty\windows\caffe3rdparty\lib\boost_system-vc140-mt-1_61.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffehdf5.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffehdf5_hl.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffezlib.lib;..\..\3rdparty\windows\caffe3rdparty\lib\gflags.lib;..\..\3rdparty\windows\caffe3rdparty\lib\glog.lib;..\..\3rdparty\windows\caffe3rdparty\lib\leveldb.lib;..\..\3rdparty\windows\caffe3rdparty\lib\libopenblas.dll.a;..\..\3rdparty\windows\caffe3rdparty\lib\libprotobuf.lib;..\..\3rdparty\windows\caffe3rdparty\lib\lmdb.lib;..\..\3rdparty\windows\caffe3rdparty\lib\snappy.lib;..\..\3rdparty\windows\opencv\x64\vc14\lib\opencv_world310.lib;cublas.lib;cublas_device.lib;cudart.lib;cudnn.lib;curand.lib;ntdll.lib;shlwapi.lib;..\x64\Release\OpenPose.lib;%(AdditionalDependencies)</AdditionalDependencies>
+      <AdditionalDependencies>..\..\3rdparty\windows\caffe3rdparty\lib\gflags.lib;..\..\3rdparty\windows\opencv\x64\vc14\lib\opencv_world310.lib;..\x64\Release\OpenPose.lib;%(AdditionalDependencies)</AdditionalDependencies>
    </Link>
  </ItemDefinitionGroup>
  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />

--- a/windows/TutorialWrapper/3_user_asynchronous.vcxproj
+++ b/windows/TutorialWrapper/3_user_asynchronous.vcxproj
@@ -72,7 +72,7 @@
      <FunctionLevelLinking>true</FunctionLevelLinking>
      <IntrinsicFunctions>true</IntrinsicFunctions>
      <SDLCheck>true</SDLCheck>
-      <AdditionalIncludeDirectories>..\..\include;..\..\3rdparty\windows\opencv\include;..\..\3rdparty\windows\caffe\include;..\..\3rdparty\windows\caffe\include2;..\..\3rdparty\windows\caffe3rdparty\include;..\..\3rdparty\windows\caffe3rdparty\include\boost-1_61;$(CUDA_PATH_V8_0)\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+      <AdditionalIncludeDirectories>..\..\include;..\..\3rdparty\windows\opencv\include;..\..\3rdparty\windows\caffe3rdparty\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
      <PreprocessorDefinitions>_CRT_SECURE_NO_WARNINGS;CMAKE_WINDOWS_BUILD;GLOG_NO_ABBREVIATED_SEVERITIES;GOOGLE_GLOG_DLL_DECL=__declspec(dllimport);GOOGLE_GLOG_DLL_DECL_FOR_UNITTESTS=__declspec(dllimport);CMAKE_INTDIR="Release";NDEBUG</PreprocessorDefinitions>
      <ObjectFileName>$(IntDir)\obj\relDir\relDir\%(RelativeDir)\%(Filename)%(Extension).obj</ObjectFileName>
      <MultiProcessorCompilation>true</MultiProcessorCompilation>
@@ -83,7 +83,7 @@
      <EnableCOMDATFolding>true</EnableCOMDATFolding>
      <OptimizeReferences>true</OptimizeReferences>
      <AdditionalLibraryDirectories>$(CUDA_PATH_V8_0)\lib\x64\;</AdditionalLibraryDirectories>
-      <AdditionalDependencies>..\..\3rdparty\windows\caffe\lib\caffe.lib;..\..\3rdparty\windows\caffe\lib\caffeproto.lib;..\..\3rdparty\windows\caffe3rdparty\lib\boost_filesystem-vc140-mt-1_61.lib;..\..\3rdparty\windows\caffe3rdparty\lib\boost_system-vc140-mt-1_61.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffehdf5.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffehdf5_hl.lib;..\..\3rdparty\windows\caffe3rdparty\lib\caffezlib.lib;..\..\3rdparty\windows\caffe3rdparty\lib\gflags.lib;..\..\3rdparty\windows\caffe3rdparty\lib\glog.lib;..\..\3rdparty\windows\caffe3rdparty\lib\leveldb.lib;..\..\3rdparty\windows\caffe3rdparty\lib\libopenblas.dll.a;..\..\3rdparty\windows\caffe3rdparty\lib\libprotobuf.lib;..\..\3rdparty\windows\caffe3rdparty\lib\lmdb.lib;..\..\3rdparty\windows\caffe3rdparty\lib\snappy.lib;..\..\3rdparty\windows\opencv\x64\vc14\lib\opencv_world310.lib;cublas.lib;cublas_device.lib;cudart.lib;cudnn.lib;curand.lib;ntdll.lib;shlwapi.lib;..\x64\Release\OpenPose.lib;%(AdditionalDependencies)</AdditionalDependencies>
+      <AdditionalDependencies>..\..\3rdparty\windows\caffe3rdparty\lib\gflags.lib;..\..\3rdparty\windows\opencv\x64\vc14\lib\opencv_world310.lib;..\x64\Release\OpenPose.lib;%(AdditionalDependencies)</AdditionalDependencies>
    </Link>
  </ItemDefinitionGroup>
  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />