diff --git a/.gitignore b/.gitignore
index 1f636f2..5ae33af 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,2 +1,4 @@
 include/libsgm_config.h
-build/
\ No newline at end of file
+build/
+.cache
+dist
diff --git a/CMakeLists.txt b/CMakeLists.txt
index e6a6e0a..a730f07 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -4,6 +4,7 @@ option(ENABLE_ZED_DEMO      "Build a Demo using ZED Camera" OFF)
 option(ENABLE_SAMPLES       "Build samples" OFF)
 option(ENABLE_TESTS         "Test library" OFF)
 option(LIBSGM_SHARED        "Build a shared library" OFF)
+option(BUILD_PYTHON_WRAPPER "Build pybind11 wrappers" OFF)
 option(BUILD_OPENCV_WRAPPER "Make library compatible with cv::Mat and cv::cuda::GpuMat of OpenCV" OFF)
 set(CUDA_ARCHS "52;61;72;75;86" CACHE STRING "List of architectures to generate device code for")
 
@@ -16,6 +17,14 @@ ${PROJECT_SOURCE_DIR}/include/libsgm_config.h
 
 add_subdirectory(src)
 
+if(BUILD_PYTHON_WRAPPER)
+    if (LIBSGM_SHARED)
+        add_subdirectory(pysgm)
+    else()
+        message(WARNING "Python wrappers requires LIBSGM_SHARED=ON")
+    endif()
+endif()
+
 if(ENABLE_SAMPLES)
 	add_subdirectory(sample)
 endif()
diff --git a/README.md b/README.md
index e77534e..9b79e79 100644
--- a/README.md
+++ b/README.md
@@ -37,6 +37,7 @@ The libSGM performance obtained from benchmark sample
 |OpenCV|version >= 3.4.8|for samples|
 |OpenCV CUDA module|version >= 3.4.8|for OpenCV wrapper|
 |ZED SDK|version >= 3.0|for ZED sample|
+|Pybind11|latest|for python bindings|
 
 ## Build Instructions
 ```
diff --git a/pyproject.toml b/pyproject.toml
new file mode 100644
index 0000000..786f345
--- /dev/null
+++ b/pyproject.toml
@@ -0,0 +1,26 @@
+[build-system]
+requires = ["scikit-build-core", "pybind11"]
+build-backend = "scikit_build_core.build"
+
+[project]
+name = "pysgm"
+authors = [
+	{name = "Oleksandr Slovak"},
+	{name = "Christoph Liebender", email = "christoph.liebender@uni-wuerzburg.de"}
+]
+license = {file = "LICENSE"}
+readme  = "README.md"
+version = "3.1.0"
+
+[project.urls]
+Repository = "https://github.com/fixstars/libSGM.git"
+
+[tool.scikit-build]
+cmake.args = [
+	"-DLIBSGM_SHARED=ON",
+	"-DBUILD_PYTHON_WRAPPER=ON"
+]
+wheel.exclude = [
+	"**/*.h",
+	"**/*.cmake"
+]
diff --git a/pysgm/CMakeLists.txt b/pysgm/CMakeLists.txt
new file mode 100644
index 0000000..8123fc9
--- /dev/null
+++ b/pysgm/CMakeLists.txt
@@ -0,0 +1,15 @@
+find_package(Python COMPONENTS Development Interpreter)
+find_package(pybind11 CONFIG)
+
+pybind11_add_module(pysgm MODULE "pysgm.cpp")
+target_include_directories(pysgm PRIVATE "${PROJECT_SOURCE_DIR}/include")
+target_link_libraries(pysgm PRIVATE sgm)
+
+if(BUILD_OPENCV_WRAPPER)
+    target_link_libraries(pysgm PRIVATE ${OpenCV_LIBS})
+endif()
+
+install(
+    TARGETS pysgm
+    LIBRARY DESTINATION ${CMAKE_INSTALL_PREFIX}/lib/python${Python_VERSION_MAJOR}.${Python_VERSION_MINOR}/site-packages
+)
diff --git a/pysgm/pysgm.cpp b/pysgm/pysgm.cpp
new file mode 100644
index 0000000..06f0c41
--- /dev/null
+++ b/pysgm/pysgm.cpp
@@ -0,0 +1,192 @@
+#include <pybind11/pybind11.h>
+
+#include <libsgm.h>
+#include <libsgm_wrapper.h>
+
+#ifdef BUILD_OPENCV_WRAPPER
+
+#include <pybind11/numpy.h>
+
+namespace pybind11 {
+namespace detail {
+template<>
+struct type_caster<cv::Mat> {
+ public:
+ PYBIND11_TYPE_CASTER(cv::Mat, _("numpy.ndarray"));
+
+  //! 1. cast numpy.ndarray to cv::Mat
+  bool load(handle obj, bool) {
+    array b = reinterpret_borrow<array>(obj);
+    buffer_info info = b.request();
+
+    //const int ndims = (int)info.ndim;
+    int nh = 1;
+    int nw = 1;
+    int nc = 1;
+    int ndims = info.ndim;
+    if (ndims == 2) {
+      nh = info.shape[0];
+      nw = info.shape[1];
+    } else if (ndims == 3) {
+      nh = info.shape[0];
+      nw = info.shape[1];
+      nc = info.shape[2];
+    } else {
+      char msg[64];
+      std::sprintf(msg, "Unsupported dim %d, only support 2d, or 3-d", ndims);
+      throw std::logic_error(msg);
+      return false;
+    }
+
+    int dtype;
+    if (info.format == format_descriptor<unsigned char>::format()) {
+      dtype = CV_8UC(nc);
+    } else if (info.format == format_descriptor<unsigned short>::format()) {
+      dtype = CV_16UC(nc);
+    } else if (info.format == format_descriptor<int>::format()) {
+      dtype = CV_32SC(nc);
+    } else if (info.format == format_descriptor<float>::format()) {
+      dtype = CV_32FC(nc);
+    } else {
+      throw std::logic_error("Unsupported type, only support uchar, int32, float");
+      return false;
+    }
+
+    value = cv::Mat(nh, nw, dtype, info.ptr);
+    return true;
+  }
+
+  //! 2. cast cv::Mat to numpy.ndarray
+  static handle cast(const cv::Mat &mat, return_value_policy, handle defval) {
+//    UNUSED(defval);
+
+
+    std::string format = format_descriptor<unsigned char>::format();
+    size_t elemsize = sizeof(unsigned char);
+    int nw = mat.cols;
+    int nh = mat.rows;
+    int nc = mat.channels();
+    int depth = mat.depth();
+    int type = mat.type();
+    int dim = (depth == type) ? 2 : 3;
+
+    if (depth == CV_8U) {
+      format = format_descriptor<unsigned char>::format();
+      elemsize = sizeof(unsigned char);
+    } else if (depth == CV_16U) {
+      format = format_descriptor<unsigned short>::format();
+      elemsize = sizeof(unsigned short);
+    } else if (depth == CV_16S) {
+      format = format_descriptor<short>::format();
+      elemsize = sizeof(short);
+    } else if (depth == CV_32S) {
+      format = format_descriptor<int>::format();
+      elemsize = sizeof(int);
+    } else if (depth == CV_32F) {
+      format = format_descriptor<float>::format();
+      elemsize = sizeof(float);
+    } else {
+      throw std::logic_error("Unsupport type!");
+    }
+
+    std::vector<size_t> bufferdim;
+    std::vector<size_t> strides;
+    if (dim == 2) {
+      bufferdim = {(size_t) nh, (size_t) nw};
+      strides = {elemsize * (size_t) nw, elemsize};
+    } else if (dim == 3) {
+      bufferdim = {(size_t) nh, (size_t) nw, (size_t) nc};
+      strides = {(size_t) elemsize * nw * nc, (size_t) elemsize * nc, (size_t) elemsize};
+    }
+    return array(buffer_info(mat.data, elemsize, format, dim, bufferdim, strides)).release();
+  }
+};
+}
+}//! end namespace pybind11::detail
+
+#endif // BUILD_OPENCV_WRAPPER
+
+
+#define RW(type_name, field_name) .def_readwrite(#field_name, &type_name::field_name)
+namespace py = pybind11;
+
+PYBIND11_MODULE(pysgm, m) {
+
+  m.doc() = "libSGM python binding";
+
+  py::enum_<sgm::ExecuteInOut>(m, "EXECUTE_INOUT")
+      .value("EXECUTE_INOUT_HOST2HOST", sgm::ExecuteInOut::EXECUTE_INOUT_HOST2HOST)
+      .value("EXECUTE_INOUT_HOST2CUDA", sgm::ExecuteInOut::EXECUTE_INOUT_HOST2CUDA)
+      .value("EXECUTE_INOUT_CUDA2HOST", sgm::ExecuteInOut::EXECUTE_INOUT_CUDA2HOST)
+      .value("EXECUTE_INOUT_CUDA2CUDA", sgm::ExecuteInOut::EXECUTE_INOUT_CUDA2CUDA)
+      ;
+
+  py::enum_<sgm::PathType>(m, "PathType")
+      .value("SCAN_4PATH", sgm::PathType::SCAN_4PATH)
+      .value("SCAN_8PATH", sgm::PathType::SCAN_8PATH)
+      ;
+
+  py::class_<sgm::StereoSGM> StereoSGM(m, "StereoSGM");
+
+  py::class_<sgm::StereoSGM::Parameters>(StereoSGM, "Parameters")
+      .def(py::init<int, int, float, bool, sgm::PathType, int, int>(),
+           py::arg("P1") = 10,
+           py::arg("P2") = 120,
+           py::arg("uniqueness") = 0.95f,
+           py::arg("subpixel") = false,
+           py::arg("PathType") = sgm::PathType::SCAN_8PATH,
+           py::arg("min_disp") = 0,
+           py::arg("LR_max_diff") = 1
+      )
+          RW(sgm::StereoSGM::Parameters, P1)
+          RW(sgm::StereoSGM::Parameters, P2)
+          RW(sgm::StereoSGM::Parameters, uniqueness)
+          RW(sgm::StereoSGM::Parameters, subpixel)
+          RW(sgm::StereoSGM::Parameters, path_type)
+          RW(sgm::StereoSGM::Parameters, min_disp)
+          RW(sgm::StereoSGM::Parameters, LR_max_diff);
+
+  StereoSGM
+      .def(py::init<int, int, int, int, int, sgm::ExecuteInOut, const sgm::StereoSGM::Parameters &>(),
+           py::arg("width") = 612,
+           py::arg("height") = 514,
+           py::arg("disparity_size") = 128,
+           py::arg("input_depth_bits") = 8U,
+           py::arg("output_depth_bits") = 8U,
+           py::arg("inout_type") = sgm::ExecuteInOut::EXECUTE_INOUT_HOST2HOST,
+           py::arg("param") = sgm::StereoSGM::Parameters()
+               )
+      .def(py::init<int, int, int, int, int, int, int, sgm::ExecuteInOut, const sgm::StereoSGM::Parameters &>())
+      .def("execute", [](sgm::StereoSGM &w, uintptr_t left_pixels, uintptr_t right_pixels, uintptr_t dst) {
+        w.execute((void *)left_pixels, (void *)right_pixels, (void *)dst);
+      })
+      .def("get_invalid_disparity", &sgm::StereoSGM::get_invalid_disparity)
+      ;
+
+#ifdef BUILD_OPENCV_WRAPPER
+
+  py::class_<sgm::LibSGMWrapper> LibSGMWrapper(m, "LibSGMWrapper");
+
+  LibSGMWrapper
+      .def(py::init<int, int, int, float, bool, sgm::PathType, int, int>(),
+           py::arg("numDisparity") = 128,
+           py::arg("P1") = 10,
+           py::arg("P2") = 120,
+           py::arg("uniquenessRatio") = 0.95f,
+           py::arg("subpixel") = false,
+           py::arg("pathType") = sgm::PathType::SCAN_8PATH,
+           py::arg("minDisparity") = 0,
+           py::arg("lrMaxDiff") = 1)
+      .def("getInvalidDisparity", &sgm::LibSGMWrapper::getInvalidDisparity)
+      .def("hasSubpixel", &sgm::LibSGMWrapper::hasSubpixel)
+      .def("execute", [](sgm::LibSGMWrapper &w, cv::Mat &left_pixels, const cv::Mat &right_pixels) {
+        cv::Mat disp;
+        w.execute(left_pixels, right_pixels, disp);
+        return disp;
+      });
+
+#endif // BUILD_OPENCV_WRAPPER
+
+  m.def("SUBPIXEL_SCALE", []() { return sgm::StereoSGM::SUBPIXEL_SCALE; });
+  m.def("SUBPIXEL_SHIFT", []() { return sgm::StereoSGM::SUBPIXEL_SHIFT; });
+}
diff --git a/sample/pysgm/pysgm_test_opencv_wrap.py b/sample/pysgm/pysgm_test_opencv_wrap.py
new file mode 100644
index 0000000..52b2ed9
--- /dev/null
+++ b/sample/pysgm/pysgm_test_opencv_wrap.py
@@ -0,0 +1,36 @@
+#!/usr/bin/env python3
+
+import sys
+
+import cv2
+import numpy as np
+
+import pysgm
+
+disp_size = 128
+
+sgm_opencv_wrapper = pysgm.LibSGMWrapper(
+    numDisparity=int(disp_size), P1=int(10), P2=int(120), uniquenessRatio=np.float32(0.95),
+    subpixel=False, pathType=pysgm.PathType.SCAN_8PATH, minDisparity=int(0),
+    lrMaxDiff=int(1))
+
+
+I1 = cv2.imread(sys.argv[1], cv2.IMREAD_GRAYSCALE)
+I2 = cv2.imread(sys.argv[2], cv2.IMREAD_GRAYSCALE)
+
+disp = sgm_opencv_wrapper.execute(I1, I2)
+
+if sgm_opencv_wrapper.hasSubpixel():
+    disp = disp.astype('float') / pysgm.SUBPIXEL_SCALE()
+
+if sgm_opencv_wrapper.hasSubpixel():
+    mask = disp == (sgm_opencv_wrapper.getInvalidDisparity() / pysgm.SUBPIXEL_SCALE())
+else:
+    mask = disp == sgm_opencv_wrapper.getInvalidDisparity()
+
+disp = (255. * disp / disp_size)
+disp_color = cv2.applyColorMap(disp.astype("uint8"), cv2.COLORMAP_JET)
+disp_color[mask, :] = 0
+
+cv2.imshow("disp_color", disp_color)
+cv2.waitKey(0)
diff --git a/sample/pysgm/pysgm_test_raw.py b/sample/pysgm/pysgm_test_raw.py
new file mode 100644
index 0000000..3ffce1d
--- /dev/null
+++ b/sample/pysgm/pysgm_test_raw.py
@@ -0,0 +1,39 @@
+#!/usr/bin/env python3
+
+import sys
+
+import cv2
+import numpy as np
+
+import pysgm
+
+disp_size = 128
+
+I1 = cv2.imread(sys.argv[1], cv2.IMREAD_GRAYSCALE)
+I2 = cv2.imread(sys.argv[2], cv2.IMREAD_GRAYSCALE)
+
+disp = np.zeros_like(I2)
+
+I1_ptr, _ = I1.__array_interface__['data']
+I2_ptr, _ = I2.__array_interface__['data']
+disp_ptr, _ = disp.__array_interface__['data']
+
+height, width = I1.shape
+
+params = pysgm.StereoSGM.Parameters(P1=int(10), P2=int(120), uniqueness=np.float32(0.95), subpixel=False,
+                                    PathType=pysgm.PathType.SCAN_8PATH, min_disp=int(0), LR_max_diff=int(1))
+
+sgm = pysgm.StereoSGM(width=width, height=height, disparity_size=int(disp_size), input_depth_bits=int(8),
+                      output_depth_bits=int(8), inout_type=pysgm.EXECUTE_INOUT.EXECUTE_INOUT_HOST2HOST,
+                      param=params)
+
+sgm.execute(I1_ptr, I2_ptr, disp_ptr)
+
+mask = disp == np.int16(sgm.get_invalid_disparity())
+disp = (255. * disp / disp_size)
+
+disp_color = cv2.applyColorMap(disp.astype("uint8"), cv2.COLORMAP_JET)
+disp_color[mask] = 0
+
+cv2.imshow("disp_color", disp_color)
+cv2.waitKey(0)