potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-29 23:33:05 +04:00
Code Packages Releases Wiki Activity

Merge branch 4.x

Browse Source
This commit is contained in:
Alexander Smorkalov
2024-07-01 15:15:08 +03:00
parent 12b8ed1443 39a7b3d186
commit 3abd9f2a28
62 changed files with 2178 additions and 243 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.github/PULL_REQUEST_TEMPLATE.md
+2 -2
View File
@@ -2,8 +2,8 @@
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [ ] I agree to contribute to the project under Apache 2 License.
- [ ] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [ ] The PR is proposed to the proper branch
- [ ] There is a reference to the original bug report and related work
- [ ] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
3rdparty/carotene/CMakeLists.txt
Vendored
+4
View File
@@ -52,5 +52,9 @@ if(WITH_NEON)
endif()
endif()
if(MINGW)
target_compile_definitions(carotene_objs PRIVATE "-D_USE_MATH_DEFINES=1")
endif()
# we add dummy file to fix XCode build
add_library(carotene STATIC ${OPENCV_3RDPARTY_EXCLUDE_FROM_ALL} "$<TARGET_OBJECTS:carotene_objs>" "${CAROTENE_SOURCE_DIR}/dummy.cpp")
3rdparty/hal_rvv/hal_rvv.hpp
Vendored
-2
View File
@@ -5,8 +5,6 @@
#ifndef OPENCV_HAL_RVV_HPP_INCLUDED
#define OPENCV_HAL_RVV_HPP_INCLUDED
#include <riscv_vector.h>
#include "opencv2/core/hal/interface.h"
#ifndef CV_HAL_RVV_071_ENABLED
3rdparty/hal_rvv/version/hal_rvv_071.hpp
Vendored
+2
View File
@@ -5,6 +5,8 @@
#ifndef OPENCV_HAL_RVV_071_HPP_INCLUDED
#define OPENCV_HAL_RVV_071_HPP_INCLUDED
#include <riscv_vector.h>
#include <limits>
namespace cv { namespace cv_hal_rvv {
3rdparty/ippicv/CMakeLists.txt
Vendored
+1 -1
View File
@@ -10,7 +10,7 @@ if(HAVE_IPP_ICV)
add_definitions(-DICV_BASE)
endif()
file(GLOB lib_srcs ${IPP_IW_PATH}/src/*.c)
file(GLOB lib_srcs ${IPP_IW_PATH}/src/*.c ${IPP_IW_PATH}/src/*.cpp)
file(GLOB lib_hdrs ${IPP_IW_PATH}/include/*.h ${IPP_IW_PATH}/include/iw/*.h ${IPP_IW_PATH}/include/iw++/*.hpp)
# ----------------------------------------------------------------------------------
CMakeLists.txt
+15
View File
@@ -307,6 +307,10 @@ OCV_OPTION(WITH_GTK "Include GTK support" ON
OCV_OPTION(WITH_GTK_2_X "Use GTK version 2" OFF
VISIBLE_IF UNIX AND NOT APPLE AND NOT ANDROID
VERIFY HAVE_GTK AND NOT HAVE_GTK3)
OCV_OPTION(WITH_FRAMEBUFFER "Include framebuffer support" OFF
VISIBLE_IF UNIX AND NOT APPLE AND NOT ANDROID)
OCV_OPTION(WITH_FRAMEBUFFER_XVFB "Include virtual framebuffer support" OFF
VISIBLE_IF UNIX AND NOT APPLE AND NOT ANDROID)
OCV_OPTION(WITH_WAYLAND "Include Wayland support" OFF
VISIBLE_IF UNIX AND NOT APPLE AND NOT ANDROID
VERIFY HAVE_WAYLAND)
@@ -1019,9 +1023,13 @@ foreach(hal ${OpenCV_HAL})
message(STATUS "NDSRVP: Andes GNU Toolchain DSP extension is not open, disabling ndsrvp...")
endif()
elseif(hal STREQUAL "halrvv")
if(";${CPU_BASELINE_FINAL};" MATCHES ";RVV;")
add_subdirectory(3rdparty/hal_rvv/)
ocv_hal_register(RVV_HAL_LIBRARIES RVV_HAL_HEADERS RVV_HAL_INCLUDE_DIRS)
list(APPEND OpenCV_USED_HAL "HAL RVV (ver ${RVV_HAL_VERSION})")
else()
message(STATUS "HAL RVV: RVV is not available, disabling halrvv...")
endif()
else()
ocv_debug_message(STATUS "OpenCV HAL: ${hal} ...")
ocv_clear_vars(OpenCV_HAL_LIBRARIES OpenCV_HAL_HEADERS OpenCV_HAL_INCLUDE_DIRS)
@@ -1444,6 +1452,13 @@ if(WITH_GTK OR HAVE_GTK)
endif()
endif()
if(WITH_FRAMEBUFFER OR HAVE_FRAMEBUFFER)
status(" Framebuffer UI:" HAVE_FRAMEBUFFER THEN YES ELSE NO)
if(WITH_FRAMEBUFFER_XVFB OR HAVE_FRAMEBUFFER_XVFB)
status(" Virtual framebuffer UI:" HAVE_FRAMEBUFFER_XVFB THEN YES ELSE NO)
endif()
endif()
if(WITH_OPENGL OR HAVE_OPENGL)
status(" OpenGL support:" HAVE_OPENGL THEN "YES (${OPENGL_LIBRARIES})" ELSE NO)
endif()
cmake/OpenCVFindIPP.cmake
+16 -3
View File
@@ -87,9 +87,14 @@ macro(ipp_detect_version)
get_filename_component(IPP_INCLUDE_DIRS ${IPP_VERSION_FILE} PATH)
set(__msg)
set(IPP_NEW_LAYOUT 0)
if(EXISTS ${IPP_ROOT_DIR}/include/ippicv_redefs.h)
set(__msg " (ICV version)")
set(HAVE_IPP_ICV 1)
elseif(EXISTS ${IPP_ROOT_DIR}/include/ipp/ipp.h)
set(IPP_NEW_LAYOUT 1)
# workaround to enable both layouts
add_definitions(-DIPP_PRESERVE_OLD_LAYOUT)
elseif(EXISTS ${IPP_ROOT_DIR}/include/ipp.h)
# nothing
else()
@@ -117,10 +122,18 @@ macro(ipp_detect_version)
if(APPLE AND NOT HAVE_IPP_ICV)
_ipp_set_library_dir(${IPP_ROOT_DIR}/lib)
elseif(IPP_X64)
_ipp_set_library_dir(${IPP_ROOT_DIR}/lib/intel64)
elseif (IPP_NEW_LAYOUT)
if(IPP_X64)
_ipp_set_library_dir(${IPP_ROOT_DIR}/lib)
else()
_ipp_set_library_dir(${IPP_ROOT_DIR}/lib32)
endif()
else()
_ipp_set_library_dir(${IPP_ROOT_DIR}/lib/ia32)
if(IPP_X64)
_ipp_set_library_dir(${IPP_ROOT_DIR}/lib/intel64)
else()
_ipp_set_library_dir(${IPP_ROOT_DIR}/lib/ia32)
endif()
endif()
macro(_ipp_add_library name)
cmake/OpenCVUtils.cmake
+1 -1
View File
@@ -1992,7 +1992,7 @@ macro(ocv_git_describe var_name path)
OUTPUT_STRIP_TRAILING_WHITESPACE
)
if(NOT GIT_RESULT EQUAL 0)
execute_process(COMMAND "${GIT_EXECUTABLE}" describe --tags --always --dirty --match "[0-9].[0-9].[0-9]*" --exclude "[^-]*-cvsdk"
execute_process(COMMAND "${GIT_EXECUTABLE}" describe --tags --always --dirty --match "[0-9].[0-9]*.[0-9]*" --exclude "[^-]*-cvsdk"
WORKING_DIRECTORY "${path}"
OUTPUT_VARIABLE ${var_name}
RESULT_VARIABLE GIT_RESULT
cmake/checks/framebuffer.cpp
+9
View File
@@ -0,0 +1,9 @@
#include <X11/XWDFile.h>
#include <X11/X.h>
int main(void)
{
XWDFileHeader *xwd_header;
XWDColor *xwd_colors;
return 0;
}
doc/pattern_tools/svgfig.py
Executable → Regular
+49 -26
View File
@@ -1,20 +1,32 @@
# svgfig.py copyright (C) 2008 Jim Pivarski <jpivarski@gmail.com>
#
# This program is free software; you can redistribute it and/or
# modify it under the terms of the GNU General Public License
# as published by the Free Software Foundation; either version 2
# of the License, or (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program; if not, write to the Free Software
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA
#
# Full licence is in the file COPYING and at http://www.gnu.org/copyleft/gpl.html
# BSD 3-Clause License
# Copyright (c) 2022, Jim Pivarski
# All rights reserved.
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are met:
# 1. Redistributions of source code must retain the above copyright notice, this
# list of conditions and the following disclaimer.
# 2. Redistributions in binary form must reproduce the above copyright notice,
# this list of conditions and the following disclaimer in the documentation
# and/or other materials provided with the distribution.
# 3. Neither the name of the copyright holder nor the names of its
# contributors may be used to endorse or promote products derived from
# this software without specific prior written permission.
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
# DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
# FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
# DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
# SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
# CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
# OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
import re, codecs, os, platform, copy, itertools, math, cmath, random, sys, copy
_epsilon = 1e-5
@@ -29,16 +41,24 @@ try:
except NameError:
unicode = lambda s: str(s)
try:
xrange # Python 2
except NameError:
xrange = range # Python 3
if re.search("windows", platform.system(), re.I):
try:
import _winreg
_default_directory = _winreg.QueryValueEx(_winreg.OpenKey(_winreg.HKEY_CURRENT_USER,
r"Software\Microsoft\Windows\Current Version\Explorer\Shell Folders"), "Desktop")[0]
# tmpdir = _winreg.QueryValueEx(_winreg.OpenKey(_winreg.HKEY_CURRENT_USER, "Environment"), "TEMP")[0]
# if tmpdir[0:13] != "%USERPROFILE%":
# tmpdir = os.path.expanduser("~") + tmpdir[13:]
except:
_default_directory = os.path.expanduser("~") + os.sep + "Desktop"
_default_fileName = "tmp.svg"
_hacks = {}
_hacks["inkscape-text-vertical-shift"] = False
__version__ = "1.0.1"
def rgb(r, g, b, maximum=1.):
"""Create an SVG color string "#xxyyzz" from r, g, and b.
@@ -437,9 +457,12 @@ class SVG:
return output
@staticmethod
def interpret_fileName(fileName=None):
return fileName or _default_fileName
def interpret_fileName(self, fileName=None):
if fileName is None:
fileName = _default_fileName
if re.search("windows", platform.system(), re.I) and not os.path.isabs(fileName):
fileName = _default_directory + os.sep + fileName
return fileName
def save(self, fileName=None, encoding="utf-8", compresslevel=None):
"""Save to a file for viewing. Note that svg.save() overwrites the file named _default_fileName.
@@ -591,7 +614,7 @@ def template(fileName, svg, replaceme="REPLACEME"):
def load(fileName):
"""Loads an SVG image from a file."""
return load_stream(open(fileName))
return load_stream(file(fileName))
def load_stream(stream):
"""Loads an SVG image from a stream (can be a string or a file object)."""
@@ -1848,7 +1871,7 @@ class Poly:
piecewise-linear segments joining the (x,y) points
"bezier"/"B" d=[(x, y, c1x, c1y, c2x, c2y), ...]
Bezier curve with two control points (control points
precede (x,y), as in SVG paths). If (c1x,c1y) and
preceed (x,y), as in SVG paths). If (c1x,c1y) and
(c2x,c2y) both equal (x,y), you get a linear
interpolation ("lines")
"velocity"/"V" d=[(x, y, vx, vy), ...]
doc/py_tutorials/py_imgproc/py_template_matching/py_template_matching.markdown
+3 -3
View File
@@ -46,12 +46,12 @@ assert template is not None, "file could not be read, check with os.path.exists(
w, h = template.shape[::-1]
# All the 6 methods for comparison in a list
methods = ['cv.TM_CCOEFF', 'cv.TM_CCOEFF_NORMED', 'cv.TM_CCORR',
'cv.TM_CCORR_NORMED', 'cv.TM_SQDIFF', 'cv.TM_SQDIFF_NORMED']
methods = ['TM_CCOEFF', 'TM_CCOEFF_NORMED', 'TM_CCORR',
'TM_CCORR_NORMED', 'TM_SQDIFF', 'TM_SQDIFF_NORMED']
for meth in methods:
img = img2.copy()
method = eval(meth)
method = getattr(cv, meth)
# Apply template Matching
res = cv.matchTemplate(img,template,method)
doc/tutorials/app/intelperc.markdown
+1 -1
View File
@@ -3,7 +3,7 @@ Using Creative Senz3D and other Intel RealSense SDK compatible depth sensors {#t
@tableofcontents
@prev_tutorial{tutorial_orbbec_astra}
@prev_tutorial{tutorial_orbbec_astra_openni}
@next_tutorial{tutorial_wayland_ubuntu}
![hardwares](images/realsense.jpg)
doc/tutorials/app/kinect_openni.markdown
+1 -1
View File
@@ -4,7 +4,7 @@ Using Kinect and other OpenNI compatible depth sensors {#tutorial_kinect_openni}
@tableofcontents
@prev_tutorial{tutorial_video_write}
@next_tutorial{tutorial_orbbec_astra}
@next_tutorial{tutorial_orbbec_astra_openni}
Depth sensors compatible with OpenNI (Kinect, XtionPRO, ...) are supported through VideoCapture
doc/tutorials/app/orbbec_astra.markdown → doc/tutorials/app/orbbec_astra_openni.markdown
+15 -9
View File
@@ -1,4 +1,4 @@
Using Orbbec Astra 3D cameras {#tutorial_orbbec_astra}
Using Orbbec Astra 3D cameras {#tutorial_orbbec_astra_openni}
======================================================
@tableofcontents
@@ -9,7 +9,7 @@ Using Orbbec Astra 3D cameras {#tutorial_orbbec_astra}
### Introduction
This tutorial is devoted to the Astra Series of Orbbec 3D cameras (https://orbbec3d.com/index/Product/info.html?cate=38&id=36).
This tutorial is devoted to the Astra Series of Orbbec 3D cameras (https://www.orbbec.com/products/structured-light-camera/astra-series/).
That cameras have a depth sensor in addition to a common color sensor. The depth sensors can be read using
the open source OpenNI API with @ref cv::VideoCapture class. The video stream is provided through the regular
camera interface.
@@ -18,7 +18,7 @@ camera interface.
In order to use the Astra camera's depth sensor with OpenCV you should do the following steps:
-# Download the latest version of Orbbec OpenNI SDK (from here <https://orbbec3d.com/index/download.html>).
-# Download the latest version of Orbbec OpenNI SDK (from here <https://www.orbbec.com/developers/openni-sdk/>).
Unzip the archive, choose the build according to your operating system and follow installation
steps provided in the Readme file.
@@ -70,6 +70,12 @@ In order to use the Astra camera's depth sensor with OpenCV you should do the fo
echo "exit"
@endcode
@note The last tried version `2.3.0.86_202210111154_4c8f5aa4_beta6` does not work correctly with
modern Linux, even after libusb rebuild as recommended by the instruction. The last know good
configuration is version 2.3.0.63 (tested with Ubuntu 18.04 amd64). It's not provided officialy
with the downloading page, but published by Orbbec technical suport on Orbbec community forum
[here](https://3dclub.orbbec3d.com/t/universal-download-thread-for-astra-series-cameras/622).
-# Now you can configure OpenCV with OpenNI support enabled by setting the `WITH_OPENNI2` flag in CMake.
You may also like to enable the `BUILD_EXAMPLES` flag to get a code sample working with your Astra camera.
Run the following commands in the directory containing OpenCV source code to enable OpenNI support:
@@ -106,7 +112,7 @@ can be read using the OpenNI interface with @ref cv::VideoCapture class. The vid
not available through OpenNI API and is only provided via the regular camera interface.
So, to get both depth and color frames, two @ref cv::VideoCapture objects should be created:
@snippetlineno samples/cpp/tutorial_code/videoio/orbbec_astra/orbbec_astra.cpp Open streams
@snippetlineno samples/cpp/tutorial_code/videoio/openni_orbbec_astra/openni_orbbec_astra.cpp Open streams
The first object will use the OpenNI2 API to retrieve depth data. The second one uses the
Video4Linux2 interface to access the color sensor. Note that the example above assumes that
@@ -119,12 +125,12 @@ For this example, we’ll configure width and height of both streams to VGA reso
the maximum resolution available for both sensors, and we’d like both stream parameters to be the
same for easier color-to-depth data registration:
@snippetlineno samples/cpp/tutorial_code/videoio/orbbec_astra/orbbec_astra.cpp Setup streams
@snippetlineno samples/cpp/tutorial_code/videoio/openni_orbbec_astra/openni_orbbec_astra.cpp Setup streams
For setting and retrieving some property of sensor data generators use @ref cv::VideoCapture::set and
@ref cv::VideoCapture::get methods respectively, e.g. :
@snippetlineno samples/cpp/tutorial_code/videoio/orbbec_astra/orbbec_astra.cpp Get properties
@snippetlineno samples/cpp/tutorial_code/videoio/openni_orbbec_astra/openni_orbbec_astra.cpp Get properties
The following properties of cameras available through OpenNI interface are supported for the depth
generator:
@@ -156,7 +162,7 @@ As there are two video sources that should be read simultaneously, it’s necess
threads to avoid blocking. Example implementation that gets frames from each sensor in a new thread
and stores them in a list along with their timestamps:
@snippetlineno samples/cpp/tutorial_code/videoio/orbbec_astra/orbbec_astra.cpp Read streams
@snippetlineno samples/cpp/tutorial_code/videoio/openni_orbbec_astra/openni_orbbec_astra.cpp Read streams
VideoCapture can retrieve the following data:
@@ -177,7 +183,7 @@ two video streams may become out of sync even when both streams are set up for t
A post-synchronization procedure can be applied to the streams to combine depth and color frames into
pairs. The sample code below demonstrates this procedure:
@snippetlineno samples/cpp/tutorial_code/videoio/orbbec_astra/orbbec_astra.cpp Pair frames
@snippetlineno samples/cpp/tutorial_code/videoio/openni_orbbec_astra/openni_orbbec_astra.cpp Pair frames
In the code snippet above the execution is blocked until there are some frames in both frame lists.
When there are new frames, their timestamps are being checked -- if they differ more than a half of
@@ -194,5 +200,5 @@ but the depth data makes it easy.
![Depth frame](images/astra_depth.png)
The complete implementation can be found in
[orbbec_astra.cpp](https://github.com/opencv/opencv/tree/5.x/samples/cpp/tutorial_code/videoio/orbbec_astra/orbbec_astra.cpp)
[openni_orbbec_astra.cpp](https://github.com/opencv/opencv/tree/5.x/samples/cpp/tutorial_code/videoio/openni_orbbec_astra/openni_orbbec_astra.cpp)
in `samples/cpp/tutorial_code/videoio` directory.
doc/tutorials/app/table_of_content_app.markdown
+1 -1
View File
@@ -6,6 +6,6 @@ Application utils (highgui, imgcodecs, videoio modules) {#tutorial_table_of_cont
- @subpage tutorial_video_input_psnr_ssim
- @subpage tutorial_video_write
- @subpage tutorial_kinect_openni
- @subpage tutorial_orbbec_astra
- @subpage tutorial_orbbec_astra_openni
- @subpage tutorial_intelperc
- @subpage tutorial_wayland_ubuntu
modules/core/include/opencv2/core/eigen.hpp
+22
View File
@@ -287,6 +287,17 @@ void cv2eigen( const Mat& src,
}
}
template<typename _Tp> static inline
void cv2eigen( const Mat& src,
Eigen::Matrix<_Tp, Eigen::Dynamic, Eigen::Dynamic, Eigen::RowMajor>& dst )
{
CV_CheckEQ(src.dims, 2, "");
dst.resize(src.rows, src.cols);
const Mat _dst(src.rows, src.cols, traits::Type<_Tp>::value,
dst.data(), (size_t)(dst.outerStride()*sizeof(_Tp)));
src.convertTo(_dst, _dst.type());
}
// Matx case
template<typename _Tp, int _rows, int _cols> static inline
void cv2eigen( const Matx<_Tp, _rows, _cols>& src,
@@ -307,6 +318,17 @@ void cv2eigen( const Matx<_Tp, _rows, _cols>& src,
}
}
template<typename _Tp, int _rows, int _cols> static inline
void cv2eigen( const Matx<_Tp, _rows, _cols>& src,
Eigen::Matrix<_Tp, Eigen::Dynamic, Eigen::Dynamic, Eigen::RowMajor>& dst )
{
CV_CheckEQ(src.dims, 2, "");
dst.resize(_rows, _cols);
const Mat _dst(_rows, _cols, traits::Type<_Tp>::value,
dst.data(), (size_t)(dst.outerStride()*sizeof(_Tp)));
Mat(src).copyTo(_dst);
}
template<typename _Tp> static inline
void cv2eigen( const Mat& src,
Eigen::Matrix<_Tp, Eigen::Dynamic, 1>& dst )
modules/core/src/lut.cpp
+6 -1
View File
@@ -55,6 +55,11 @@ static void LUT8u_32s( const uchar* src, const int* lut, int* dst, int len, int
LUT8u_( src, lut, dst, len, cn, lutcn );
}
static void LUT8u_16f( const uchar* src, const hfloat* lut, hfloat* dst, int len, int cn, int lutcn )
{
LUT8u_( src, lut, dst, len, cn, lutcn );
}
static void LUT8u_32f( const uchar* src, const float* lut, float* dst, int len, int cn, int lutcn )
{
LUT8u_( src, lut, dst, len, cn, lutcn );
@@ -70,7 +75,7 @@ typedef void (*LUTFunc)( const uchar* src, const uchar* lut, uchar* dst, int len
static LUTFunc lutTab[CV_DEPTH_MAX] =
{
(LUTFunc)LUT8u_8u, (LUTFunc)LUT8u_8s, (LUTFunc)LUT8u_16u, (LUTFunc)LUT8u_16s,
(LUTFunc)LUT8u_32s, (LUTFunc)LUT8u_32f, (LUTFunc)LUT8u_64f, 0
(LUTFunc)LUT8u_32s, (LUTFunc)LUT8u_32f, (LUTFunc)LUT8u_64f, (LUTFunc)LUT8u_16f
};
#ifdef HAVE_OPENCL
modules/core/src/mean.dispatch.cpp
+9 -6
View File
@@ -484,6 +484,10 @@ void meanStdDev(InputArray _src, OutputArray _mean, OutputArray _sdv, InputArray
int dcn = (int)mean_mat.total();
CV_Assert( mean_mat.type() == CV_64F && mean_mat.isContinuous() &&
(mean_mat.cols == 1 || mean_mat.rows == 1) && dcn >= cn );
double* dptr = mean_mat.ptr<double>();
for(k = cn ; k < dcn; k++ )
dptr[k] = 0;
}
if (_sdv.needed())
@@ -495,6 +499,11 @@ void meanStdDev(InputArray _src, OutputArray _mean, OutputArray _sdv, InputArray
int dcn = (int)stddev_mat.total();
CV_Assert( stddev_mat.type() == CV_64F && stddev_mat.isContinuous() &&
(stddev_mat.cols == 1 || stddev_mat.rows == 1) && dcn >= cn );
double* dptr = stddev_mat.ptr<double>();
for(k = cn ; k < dcn; k++ )
dptr[k] = 0;
}
if (src.isContinuous() && mask.isContinuous())
@@ -596,23 +605,17 @@ void meanStdDev(InputArray _src, OutputArray _mean, OutputArray _sdv, InputArray
if (_mean.needed())
{
const double* sptr = s;
int dcn = (int)mean_mat.total();
double* dptr = mean_mat.ptr<double>();
for( k = 0; k < cn; k++ )
dptr[k] = sptr[k];
for( ; k < dcn; k++ )
dptr[k] = 0;
}
if (_sdv.needed())
{
const double* sptr = sq;
int dcn = (int)stddev_mat.total();
double* dptr = stddev_mat.ptr<double>();
for( k = 0; k < cn; k++ )
dptr[k] = sptr[k];
for( ; k < dcn; k++ )
dptr[k] = 0;
}
}
modules/core/test/test_arithm.cpp
+101 -2
View File
@@ -3481,7 +3481,7 @@ TEST_P(NonZeroSupportedMatDepth, hasNonZero)
INSTANTIATE_TEST_CASE_P(
NonZero,
NonZeroSupportedMatDepth,
testing::Values(perf::MatDepth(CV_16F), CV_16BF, CV_Bool, CV_64U, CV_64S, CV_32U)
testing::Values(CV_16BF, CV_Bool, CV_64U, CV_64S, CV_32U)
);
///////////////////////////////////////////////////////////////////////////////////
@@ -3498,7 +3498,7 @@ TEST_P(LutNotSupportedMatDepth, lut)
INSTANTIATE_TEST_CASE_P(
Lut,
LutNotSupportedMatDepth,
testing::Values(perf::MatDepth(CV_16F), CV_16BF, CV_Bool, CV_64U, CV_64S, CV_32U)
testing::Values(CV_16BF, CV_Bool, CV_64U, CV_64S, CV_32U)
);
///////////////////////////////////////////////////////////////////////////////////
@@ -3526,4 +3526,103 @@ INSTANTIATE_TEST_CASE_P(
testing::Values(perf::MatDepth(CV_16F), CV_16BF, CV_Bool, CV_64U, CV_64S, CV_32U)
);
CV_ENUM(LutMatType, CV_8U, CV_16U, CV_16F, CV_32S, CV_32F, CV_64F)
struct Core_LUT: public testing::TestWithParam<LutMatType>
{
template<typename T, int ch>
cv::Mat referenceWithType(cv::Mat input, cv::Mat table)
{
cv::Mat ref(input.size(), CV_MAKE_TYPE(table.type(), ch));
for (int i = 0; i < input.rows; i++)
{
for (int j = 0; j < input.cols; j++)
{
if(ch == 1)
{
ref.at<T>(i, j) = table.at<T>(input.at<uchar>(i, j));
}
else
{
Vec<T, ch> val;
for (int k = 0; k < ch; k++)
{
val[k] = table.at<T>(input.at<Vec<uchar, ch>>(i, j)[k]);
}
ref.at<Vec<T, ch>>(i, j) = val;
}
}
}
return ref;
}
template<int ch = 1>
cv::Mat reference(cv::Mat input, cv::Mat table)
{
if (table.type() == CV_8U)
{
return referenceWithType<uchar, ch>(input, table);
}
else if (table.type() == CV_16U)
{
return referenceWithType<ushort, ch>(input, table);
}
else if (table.type() == CV_16F)
{
return referenceWithType<ushort, ch>(input, table);
}
else if (table.type() == CV_32S)
{
return referenceWithType<int, ch>(input, table);
}
else if (table.type() == CV_32F)
{
return referenceWithType<float, ch>(input, table);
}
else if (table.type() == CV_64F)
{
return referenceWithType<double, ch>(input, table);
}
return cv::Mat();
}
};
TEST_P(Core_LUT, accuracy)
{
int type = GetParam();
cv::Mat input(117, 113, CV_8UC1);
randu(input, 0, 256);
cv::Mat table(1, 256, CV_MAKE_TYPE(type, 1));
randu(table, 0, 127);
cv::Mat output;
cv::LUT(input, table, output);
cv::Mat gt = reference(input, table);
ASSERT_EQ(0, cv::norm(output, gt, cv::NORM_INF));
}
TEST_P(Core_LUT, accuracy_multi)
{
int type = (int)GetParam();
cv::Mat input(117, 113, CV_8UC3);
randu(input, 0, 256);
cv::Mat table(1, 256, CV_MAKE_TYPE(type, 1));
randu(table, 0, 127);
cv::Mat output;
cv::LUT(input, table, output);
cv::Mat gt = reference<3>(input, table);
ASSERT_EQ(0, cv::norm(output, gt, cv::NORM_INF));
}
INSTANTIATE_TEST_CASE_P(/**/, Core_LUT, LutMatType::all());
}} // namespace
modules/core/test/test_mat.cpp
+21
View File
@@ -2263,6 +2263,27 @@ TEST(Core_Eigen, eigen2cv_check_Mat_type)
EXPECT_ANY_THROW(eigen2cv(eigen_A, d_mat));
//EXPECT_EQ(CV_64FC1, d_mat.type());
}
TEST(Core_Eigen, cv2eigen_check_RowMajor)
{
Mat A(3, 2, CV_32FC1, Scalar::all(0));
A.at<float>(0,0) = 1.0;
A.at<float>(0,1) = 2.0;
A.at<float>(1,0) = 3.0;
A.at<float>(1,1) = 4.0;
A.at<float>(2,0) = 5.0;
A.at<float>(2,1) = 6.0;
Eigen::Matrix<float, Eigen::Dynamic, Eigen::Dynamic, Eigen::RowMajor> eigen_A;
EXPECT_NO_THROW(cv2eigen(A, eigen_A));
ASSERT_EQ(1.0, eigen_A(0, 0));
ASSERT_EQ(2.0, eigen_A(0, 1));
ASSERT_EQ(3.0, eigen_A(1, 0));
ASSERT_EQ(4.0, eigen_A(1, 1));
ASSERT_EQ(5.0, eigen_A(2, 0));
ASSERT_EQ(6.0, eigen_A(2, 1));
}
#endif // HAVE_EIGEN
#ifdef OPENCV_EIGEN_TENSOR_SUPPORT
modules/dnn/include/opencv2/dnn/all_layers.hpp
+10
View File
@@ -1193,6 +1193,16 @@ CV__DNN_INLINE_NS_BEGIN
static Ptr<CastLayer> create(const LayerParams &params);
};
class CV_EXPORTS DepthToSpaceLayer : public Layer {
public:
static Ptr<DepthToSpaceLayer> create(const LayerParams &params);
};
class CV_EXPORTS SpaceToDepthLayer : public Layer {
public:
static Ptr<SpaceToDepthLayer> create(const LayerParams &params);
};
//! @}
//! @}
CV__DNN_INLINE_NS_END
modules/dnn/src/cuda4dnn/csl/tensor.hpp
+12 -17
View File
@@ -271,7 +271,6 @@ namespace cv { namespace dnn { namespace cuda4dnn { namespace csl {
typename std::enable_if<cxx_utils::is_forward_iterator<ForwardItr>::value, void>
::type reshape_(ForwardItr start, ForwardItr end) {
CV_Assert(start != end);
CV_Assert(std::distance(start, end) <= rank());
using ItrValueType = typename std::iterator_traits<ForwardItr>::value_type;
@@ -290,6 +289,7 @@ namespace cv { namespace dnn { namespace cuda4dnn { namespace csl {
auto total = std::accumulate(start, end, 1, std::multiplies<ItrValueType>());
if (total < 0) {
/* there is an unknown size */
CV_CheckEQ(size() % std::abs(total), static_cast<size_type>(0), "cannot be reshaped"); // must be divisible
if (std::abs(total) <= size()) {
unknown_size = size() / std::abs(total);
total = size();
@@ -304,11 +304,9 @@ namespace cv { namespace dnn { namespace cuda4dnn { namespace csl {
CV_Error(Error::StsBadArg, "new axes do not preserve the tensor element count");
}
/* we assume the size of the unspecified axes to be one */
std::fill(std::begin(shape), std::end(shape), 1);
std::copy_backward(start, end, std::end(shape));
/* replace the unknown axis with the correct value */
/* copy shape from given iterator and reshape -1 with deduced value */
shape.resize(std::distance(start, end));
std::copy(start, end, shape.begin());
std::replace(std::begin(shape), std::end(shape), size_type(-1), unknown_size);
}
@@ -622,6 +620,7 @@ namespace cv { namespace dnn { namespace cuda4dnn { namespace csl {
auto total = std::accumulate(start, end, 1, std::multiplies<ItrValueType>());
if (total < 0) {
/* there is an unknown size */
CV_CheckEQ(size() % std::abs(total), static_cast<size_type>(0), "cannot be reshaped"); // must be divisible
if (std::abs(total) <= size()) {
unknown_size = size() / std::abs(total);
total = size();
@@ -636,11 +635,9 @@ namespace cv { namespace dnn { namespace cuda4dnn { namespace csl {
CV_Error(Error::StsBadArg, "new axes do not preserve the tensor element count");
}
/* we assume the size of the unspecified axes to be one */
std::fill(std::begin(shape), std::end(shape), 1);
std::copy_backward(start, end, std::end(shape));
/* replace the unknown axis with the correct value */
/* copy shape from given iterator and reshape -1 with deduced value */
shape.resize(std::distance(start, end));
std::copy(start, end, shape.begin());
std::replace(std::begin(shape), std::end(shape), size_type(-1), unknown_size);
}
@@ -997,7 +994,6 @@ namespace cv { namespace dnn { namespace cuda4dnn { namespace csl {
typename std::enable_if<!std::is_integral<ForwardItr>::value, void>
::type reshape_(ForwardItr start, ForwardItr end) {
CV_Assert(start != end);
CV_Assert(std::distance(start, end) <= rank());
using ItrValueType = typename std::iterator_traits<ForwardItr>::value_type;
@@ -1016,6 +1012,7 @@ namespace cv { namespace dnn { namespace cuda4dnn { namespace csl {
auto total = std::accumulate(start, end, 1, std::multiplies<ItrValueType>());
if (total < 0) {
/* there is an unknown size */
CV_CheckEQ(size() % std::abs(total), static_cast<size_type>(0), "cannot be reshaped"); // must be divisible
if (std::abs(total) <= size()) {
unknown_size = size() / std::abs(total);
total = size();
@@ -1030,11 +1027,9 @@ namespace cv { namespace dnn { namespace cuda4dnn { namespace csl {
CV_Error(Error::StsBadArg, "new axes do not preserve the tensor element count");
}
/* we assume the size of the unspecified axes to be one */
std::fill(std::begin(shape), std::end(shape), 1);
std::copy_backward(start, end, std::end(shape));
/* replace the unknown axis with the correct value */
/* copy shape from given iterator and reshape -1 with deduced value */
shape.resize(std::distance(start, end));
std::copy(start, end, shape.begin());
std::replace(std::begin(shape), std::end(shape), size_type(-1), unknown_size);
}
modules/dnn/src/cuda4dnn/primitives/depth_space_ops.hpp
+76
View File
@@ -0,0 +1,76 @@
// This file is part of OpenCV project.
// It is subject to the license terms in the LICENSE file found in the top-level directory
// of this distribution and at http://opencv.org/license.html.
#ifndef OPENCV_DNN_SRC_CUDA4DNN_PRIMITIVES_DEPTH_SPACE_OPS_HPP
#define OPENCV_DNN_SRC_CUDA4DNN_PRIMITIVES_DEPTH_SPACE_OPS_HPP
#include "../../op_cuda.hpp"
#include "../csl/stream.hpp"
#include "../csl/tensor.hpp"
#include "../csl/tensor_ops.hpp"
#include "../csl/memory.hpp"
#include "../kernels/permute.hpp"
#include <utility>
namespace cv { namespace dnn { namespace cuda4dnn {
template <class T>
class DepthSpaceOps final : public CUDABackendNode {
public:
using wrapper_type = GetCUDABackendWrapperType<T>;
DepthSpaceOps(csl::Stream stream_, const std::vector<int> &internal_shape_,
const std::vector<size_t> &permutation_)
: stream(std::move(stream_)), internal_shape(internal_shape_),
permutation(permutation_)
{
transposed_internal_shape = std::vector<int>(internal_shape.size());
for (size_t i = 0; i < permutation.size(); i++) {
transposed_internal_shape[i] = internal_shape[permutation[i]];
}
size_t num_elements = std::accumulate(internal_shape.begin(), internal_shape.end(), 1, std::multiplies<size_t>());
csl::WorkspaceBuilder builder;
builder.require<T>(num_elements);
scratch_mem_in_bytes = builder.required_workspace_size();
}
void forward(const std::vector<cv::Ptr<BackendWrapper>> &inputs,
const std::vector<cv::Ptr<BackendWrapper>> &outputs,
csl::Workspace &workspace) override {
CV_CheckEQ(inputs.size(), size_t(1), "DepthSpaceOps: only one input is accepted");
CV_CheckEQ(outputs.size(), size_t(1), "DepthSpaceOps: only one output is accepted");
auto input_wrapper = inputs.front().dynamicCast<wrapper_type>();
auto input = input_wrapper->getView();
CV_CheckEQ(input.rank(), size_t(4), "DepthSpaceOps: input needs to be 4-dimensional [N, C, H, W]");
auto output_wrapper = outputs.front().dynamicCast<wrapper_type>();
auto output = output_wrapper->getSpan();
auto ws_allocator = csl::WorkspaceAllocator(workspace);
auto transposed_internal = ws_allocator.get_tensor_span<T>(transposed_internal_shape.begin(), transposed_internal_shape.end());
// Call reshape on input so that it has the correct shape for permutation
input.reshape(internal_shape.begin(), internal_shape.end());
kernels::permute(stream, transposed_internal, input, permutation);
// Only copying is needed as output already has the expected shape
auto t = csl::TensorView<T>(transposed_internal);
csl::memcpy(output.get(), t.get(), output.size(), stream);
}
std::size_t get_workspace_memory_in_bytes() const noexcept override { return scratch_mem_in_bytes; }
private:
csl::Stream stream;
std::vector<int> internal_shape;
std::vector<size_t> permutation;
std::vector<int> transposed_internal_shape;
std::size_t scratch_mem_in_bytes;
};
}}} // namespace cv::dnn::cuda4dnn
#endif // OPENCV_DNN_SRC_CUDA4DNN_PRIMITIVES_DEPTH_SPACE_OPS_HPP
modules/dnn/src/init.cpp
+4
View File
@@ -165,6 +165,10 @@ void initializeLayerFactory()
CV_DNN_REGISTER_LAYER_CLASS(Attention, AttentionLayer);
CV_DNN_REGISTER_LAYER_CLASS(GroupNormalization, GroupNormLayer);
CV_DNN_REGISTER_LAYER_CLASS(Cast, CastLayer);
CV_DNN_REGISTER_LAYER_CLASS(DepthToSpace, DepthToSpaceLayer)
CV_DNN_REGISTER_LAYER_CLASS(SpaceToDepth, SpaceToDepthLayer)
CV_DNN_REGISTER_LAYER_CLASS(DepthToSpaceInt8, DepthToSpaceLayer)
CV_DNN_REGISTER_LAYER_CLASS(SpaceToDepthInt8, SpaceToDepthLayer)
CV_DNN_REGISTER_LAYER_CLASS(Crop, CropLayer);
CV_DNN_REGISTER_LAYER_CLASS(Eltwise, EltwiseLayer);
modules/dnn/src/layers/depth_space_ops_layer.cpp
+482
View File
@@ -0,0 +1,482 @@
// This file is part of OpenCV project.
// It is subject to the license terms in the LICENSE file found in the top-level directory
// of this distribution and at http://opencv.org/license.html.
#include "../precomp.hpp"
#include <opencv2/dnn/shape_utils.hpp>
// OpenCL backend
#ifdef HAVE_OPENCL
#include "opencl_kernels_dnn.hpp"
#endif
// OpenVINO backend
#ifdef HAVE_DNN_NGRAPH
#include "../op_inf_engine.hpp"
#include "../ie_ngraph.hpp"
#endif
// CUDA backend
#ifdef HAVE_CUDA
#include "../op_cuda.hpp"
#include "../cuda4dnn/primitives/depth_space_ops.hpp"
#endif
// CANN backend
#ifdef HAVE_CANN
#include "../op_cann.hpp"
#endif
// TIM-VX backend
#ifdef HAVE_TIMVX
#include "../op_timvx.hpp"
#endif
namespace cv { namespace dnn {
struct DepthSpaceOps {
MatShape internal_shape;
MatShape transposed_internal_shape;
std::vector<int> permutation;
#ifdef HAVE_OPENCL
UMat umat_permutation;
UMat umat_internal_strides;
UMat umat_transposed_internal_strides;
#endif
void finalize(InputArrayOfArrays inputs_arr, OutputArrayOfArrays outputs_arr) {
transposed_internal_shape = MatShape(internal_shape.size());
for (size_t i = 0; i < permutation.size(); i++) {
transposed_internal_shape[i] = internal_shape[permutation[i]];
}
#ifdef HAVE_OPENCL
umat_permutation.release();
umat_internal_strides.release();
umat_transposed_internal_strides.release();
#endif
}
void cpuCompute(const Mat &input, Mat &output) {
const auto output_shape = shape(output);
Mat tmp;
cv::transposeND(input.reshape(1, internal_shape), permutation, tmp);
tmp.reshape(1, output_shape).copyTo(output);
}
#ifdef HAVE_OPENCL
bool oclCompute(InputArrayOfArrays inputs_arr, OutputArrayOfArrays outputs_arr, OutputArrayOfArrays internals_arr) {
std::vector<UMat> inputs, outputs;
inputs_arr.getUMatVector(inputs);
outputs_arr.getUMatVector(outputs);
if (umat_permutation.empty() || umat_internal_strides.empty() || umat_transposed_internal_strides.empty()) {
Mat mat_permutation(1, permutation.size(), CV_32S, permutation.data());
mat_permutation.copyTo(umat_permutation);
std::vector<int> internal_strides(permutation.size(), 1), transposed_internal_stides(permutation.size(), 1);
for (int i = static_cast<int>(permutation.size()) - 2; i >= 0; i--) {
internal_strides[i] = internal_strides[i + 1] * internal_shape[i + 1];
transposed_internal_stides[i] = transposed_internal_stides[i + 1] * transposed_internal_shape[i + 1];
}
Mat mat_internal_strides(1, internal_strides.size(), CV_32S, internal_strides.data());
mat_internal_strides.copyTo(umat_internal_strides);
Mat mat_transposed_internal_strides(1, transposed_internal_stides.size(), CV_32S, transposed_internal_stides.data());
mat_transposed_internal_strides.copyTo(umat_transposed_internal_strides);
}
const auto output_shape = shape(outputs.front());
UMat tmp = inputs.front().reshape(1, static_cast<int>(internal_shape.size()), internal_shape.data());
bool use_half = (inputs_arr.depth() == CV_16F);
std::string permute_options = cv::format("-DDtype=%s", use_half ? "half" : "float");
ocl::Kernel permute_kernel("permute", ocl::dnn::permute_oclsrc, permute_options);
if (permute_kernel.empty()) {
return false;
}
UMat transposed_tmp(static_cast<int>(transposed_internal_shape.size()), transposed_internal_shape.data(), inputs_arr.depth());
size_t num_element = static_cast<size_t>(std::accumulate(internal_shape.begin(), internal_shape.end(), 1, std::multiplies<int>()));
permute_kernel.set(0, static_cast<int>(num_element));
permute_kernel.set(1, ocl::KernelArg::PtrReadOnly(tmp));
permute_kernel.set(2, ocl::KernelArg::PtrReadOnly(umat_permutation));
permute_kernel.set(3, ocl::KernelArg::PtrReadOnly(umat_internal_strides));
permute_kernel.set(4, ocl::KernelArg::PtrReadOnly(umat_transposed_internal_strides));
permute_kernel.set(5, static_cast<int>(permutation.size()));
permute_kernel.set(6, ocl::KernelArg::PtrWriteOnly(transposed_tmp));
if (!permute_kernel.run(1, &num_element, NULL, false)) {
return false;
}
transposed_tmp.reshape(1, static_cast<int>(output_shape.size()), output_shape.data()).copyTo(outputs.front());
return true;
}
#endif // HAVE_OPENCL
};
class DepthToSpaceLayerImpl CV_FINAL : public DepthToSpaceLayer, public DepthSpaceOps {
public:
DepthToSpaceLayerImpl(const LayerParams &params) {
setParamsFrom(params);
CV_CheckTrue(params.has("blocksize"), "DepthSpaceLayer: blocksize is required");
blocksize = params.get<int>("blocksize");
auto mode = params.get<std::string>("mode", "DCR");
if (mode == "CRD") {
is_crd = true;
permutation = {0, 1, 4, 2, 5, 3};
} else if (mode == "DCR") {
is_crd = false;
permutation = {0, 3, 4, 1, 5, 2};
} else {
CV_Error(Error::StsBadArg, cv::format("DepthToSpace: unsupported mode %s\n", mode.c_str()));
}
}
virtual bool supportBackend(int backendId) CV_OVERRIDE {
return backendId == DNN_BACKEND_OPENCV ||
backendId == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH ||
backendId == DNN_BACKEND_CUDA ||
backendId == DNN_BACKEND_CANN ||
(backendId == DNN_BACKEND_TIMVX && is_crd);
}
virtual bool getMemoryShapes(const std::vector<MatShape> &inputs,
const int requiredOutputs,
std::vector<MatShape> &outputs,
std::vector<MatShape> &internals) const CV_OVERRIDE {
CV_CheckEQ(inputs.size(), static_cast<size_t>(1), "DepthSpaceLayer: accepts only one input");
const auto &input = inputs.front();
CV_CheckEQ(input.size(), static_cast<size_t>(4), "DepthSpaceLayer: input needs to be 4-dimensional [N, C, H, W]");
int batch = input[0], input_depth = input[1], input_height = input[2], input_width = input[3];
int output_depth = -1, output_height = -1, output_width = -1;
CV_CheckEQ(input_depth % (blocksize * blocksize), 0,
"DepthSpaceLayer: requires input depth to be a multiple of (blocksize * blocksize)");
output_depth = input_depth / blocksize / blocksize;
output_height = input_height * blocksize;
output_width = input_width * blocksize;
outputs.assign(1, MatShape{batch, output_depth, output_height, output_width});
return false;
}
virtual void finalize(InputArrayOfArrays inputs_arr, OutputArrayOfArrays outputs_arr) CV_OVERRIDE {
std::vector<Mat> inputs;
inputs_arr.getMatVector(inputs);
auto input_shape = shape(inputs.front());
int batch = input_shape[0], input_depth = input_shape[1], input_height = input_shape[2], input_width = input_shape[3];
if (is_crd) {
internal_shape = MatShape{batch, input_depth / (blocksize * blocksize), blocksize, blocksize, input_height, input_width};
} else {
internal_shape = MatShape{batch, blocksize, blocksize, input_depth / (blocksize * blocksize), input_height, input_width};
}
DepthSpaceOps::finalize(inputs_arr, outputs_arr);
}
void forward(InputArrayOfArrays inputs_arr, OutputArrayOfArrays outputs_arr, OutputArrayOfArrays internals_arr) CV_OVERRIDE {
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
// TODO: support 8-bit int in permute kernel
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget) && inputs_arr.depth() != CV_8S,
DepthSpaceOps::oclCompute(inputs_arr, outputs_arr, internals_arr))
if (inputs_arr.depth() == CV_16F) {
forward_fallback(inputs_arr, outputs_arr, internals_arr);
return;
}
std::vector<Mat> inputs, outputs;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);
DepthSpaceOps::cpuCompute(inputs.front(), outputs.front());
}
#ifdef HAVE_DNN_NGRAPH
virtual Ptr<BackendNode> initNgraph(const std::vector<Ptr<BackendWrapper>> &inputs,
const std::vector<Ptr<BackendNode>> &nodes) CV_OVERRIDE {
using namespace ov::op;
auto input_node = nodes[0].dynamicCast<InfEngineNgraphNode>()->node;
std::shared_ptr<ov::Node> output_node;
if (is_crd) {
output_node = std::make_shared<v0::DepthToSpace>(input_node, v0::DepthToSpace::DepthToSpaceMode::DEPTH_FIRST, static_cast<size_t>(blocksize));
} else {
output_node = std::make_shared<v0::DepthToSpace>(input_node, v0::DepthToSpace::DepthToSpaceMode::BLOCKS_FIRST, static_cast<size_t>(blocksize));
}
return Ptr<BackendNode>(new InfEngineNgraphNode(output_node));
}
#endif // HAVE_DNN_NGRAPH
#ifdef HAVE_CUDA
Ptr<BackendNode> initCUDA(void *context_,
const std::vector<Ptr<BackendWrapper>>& inputs,
const std::vector<Ptr<BackendWrapper>>& outputs) override {
using namespace cv::dnn::cuda4dnn;
auto context = reinterpret_cast<csl::CSLContext*>(context_);
std::vector<size_t> perm(permutation.begin(), permutation.end());
return make_cuda_node<cuda4dnn::DepthSpaceOps>(preferableTarget, std::move(context->stream), internal_shape, perm);
}
#endif // HAVE_CUDA
#ifdef HAVE_CANN
virtual Ptr<BackendNode> initCann(const std::vector<Ptr<BackendWrapper>> &inputs,
const std::vector<Ptr<BackendWrapper>> &outputs,
const std::vector<Ptr<BackendNode>> &nodes) CV_OVERRIDE {
CV_CheckEQ(inputs.size(), static_cast<size_t>(1), "DepthToSpace/CANN: only accepts one input wrapper");
CV_CheckEQ(nodes.size(), static_cast<size_t>(1), "DepthToSpace/CANN: only accepts one input node");
auto input_tensor_wrapper = inputs.front().dynamicCast<CannBackendWrapper>();
auto input_tensor_desc = input_tensor_wrapper->getTensorDesc();
auto input_node = nodes.front().dynamicCast<CannBackendNode>()->getOp();
auto node = std::make_shared<ge::op::DepthToSpace>(name);
node->set_attr_block_size(blocksize);
if (is_crd) {
node->set_attr_mode("CRD");
} else {
node->set_attr_mode("DCR");
}
node->set_attr_data_format("NCHW");
node->set_input_x_by_name(*input_node, input_tensor_wrapper->name.c_str());
node->update_input_desc_x(*input_tensor_desc);
auto output_tensor_desc = std::make_shared<ge::TensorDesc>(ge::Shape(), ge::FORMAT_NCHW, ge::DT_FLOAT);
node->update_output_desc_y(*output_tensor_desc);
return Ptr<BackendNode>(new CannBackendNode(node));
}
#endif
#ifdef HAVE_TIMVX
virtual Ptr<BackendNode> initTimVX(void* timvx_info_,
const std::vector<Ptr<BackendWrapper>> &inputs,
const std::vector<Ptr<BackendWrapper>> &outputs,
bool isLast) CV_OVERRIDE {
auto info = reinterpret_cast<TimVXInfo*>(timvx_info_);
CV_Assert(info);
auto timvx_graph = info->getGraph();
CV_Assert(timvx_graph);
auto graph = timvx_graph->graph;
auto input_wrapper = inputs.front().dynamicCast<TimVXBackendWrapper>();
int input_wrapper_index = -1;
if (input_wrapper->isTensor()) {
input_wrapper_index = timvx_graph->getTensorIndex(input_wrapper->getTensor());
if (input_wrapper_index == -1) {
auto tmp = input_wrapper->getMat();
input_wrapper = std::make_shared<TimVXBackendWrapper>(tmp);
}
}
if (!input_wrapper->isTensor() || input_wrapper_index == 1) {
auto input_node_quant = Ptr<tim::vx::Quantization>(new tim::vx::Quantization(tim::vx::QuantType::ASYMMETRIC, 1.0f, 0));
input_wrapper->createTensor(graph, tim::vx::TensorAttribute::INPUT, input_node_quant);
input_wrapper_index = timvx_graph->addWrapper(input_wrapper);
}
auto output_wrapper = outputs.front().dynamicCast<TimVXBackendWrapper>();
auto output_node_quant = input_wrapper->getTensorQuantization();
if (isLast) {
auto shape_type = getShapeTypeFromMat(output_wrapper->getMat());
output_wrapper->setTensorShape(shape_type);
output_wrapper->createTensor(graph, tim::vx::TensorAttribute::OUTPUT, output_node_quant);
} else {
output_wrapper->createTensor(graph, tim::vx::TensorAttribute::TRANSIENT, output_node_quant);
}
int output_wrapper_index = timvx_graph->addWrapper(output_wrapper);
std::shared_ptr<tim::vx::Operation> timvx_node;
timvx_node = graph->CreateOperation<tim::vx::ops::DepthToSpace>(blocksize);
std::vector<int> input_wrapper_indices{input_wrapper_index}, output_wrapper_indices{output_wrapper_index};
return Ptr<BackendNode>(new TimVXBackendNode(timvx_graph, timvx_node, input_wrapper_indices, output_wrapper_indices));
}
#endif
private:
int blocksize;
bool is_crd;
};
Ptr<DepthToSpaceLayer> DepthToSpaceLayer::create(const LayerParams &params) {
return makePtr<DepthToSpaceLayerImpl>(params);
}
class SpaceToDepthLayerImpl CV_FINAL : public SpaceToDepthLayer, public DepthSpaceOps {
public:
SpaceToDepthLayerImpl(const LayerParams &params) {
setParamsFrom(params);
CV_CheckTrue(params.has("blocksize"), "SpaceToDepthLayer: blocksize is required");
blocksize = params.get<int>("blocksize");
permutation = {0, 3, 5, 1, 2, 4};
}
virtual bool supportBackend(int backendId) CV_OVERRIDE {
return backendId == DNN_BACKEND_OPENCV ||
backendId == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH ||
backendId == DNN_BACKEND_CUDA ||
backendId == DNN_BACKEND_CANN ||
(backendId == DNN_BACKEND_TIMVX);
}
virtual bool getMemoryShapes(const std::vector<MatShape> &inputs,
const int requiredOutputs,
std::vector<MatShape> &outputs,
std::vector<MatShape> &internals) const CV_OVERRIDE {
CV_CheckEQ(inputs.size(), static_cast<size_t>(1), "SpaceToDepthLayer: accepts only one input");
const auto &input = inputs.front();
CV_CheckEQ(input.size(), static_cast<size_t>(4), "SpaceToDepthLayer: input needs to be 4-dimensional [N, C, H, W]");
int batch = input[0], input_depth = input[1], input_height = input[2], input_width = input[3];
int output_depth = -1, output_height = -1, output_width = -1;
CV_CheckEQ(input_height % blocksize, 0, "SpaceToDepthLayer: requires input height to be a multiple of blocksize");
CV_CheckEQ(input_width % blocksize, 0, "SpaceToDepthLayer: requires input width to be a multiple of blocksize");
output_depth = input_depth * blocksize * blocksize;
output_height = input_height / blocksize;
output_width = input_width / blocksize;
outputs.assign(1, MatShape{batch, output_depth, output_height, output_width});
return false;
}
virtual void finalize(InputArrayOfArrays inputs_arr, OutputArrayOfArrays outputs_arr) CV_OVERRIDE {
std::vector<Mat> inputs;
inputs_arr.getMatVector(inputs);
auto input_shape = shape(inputs.front());
int batch = input_shape[0], input_depth = input_shape[1], input_height = input_shape[2], input_width = input_shape[3];
internal_shape = MatShape{batch, input_depth, input_height / blocksize, blocksize, input_width / blocksize, blocksize};
DepthSpaceOps::finalize(inputs_arr, outputs_arr);
}
void forward(InputArrayOfArrays inputs_arr, OutputArrayOfArrays outputs_arr, OutputArrayOfArrays internals_arr) CV_OVERRIDE {
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
// TODO: support 8-bit int in permute kernel
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget) && inputs_arr.depth() != CV_8S,
DepthSpaceOps::oclCompute(inputs_arr, outputs_arr, internals_arr))
if (inputs_arr.depth() == CV_16F) {
forward_fallback(inputs_arr, outputs_arr, internals_arr);
return;
}
std::vector<Mat> inputs, outputs;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);
DepthSpaceOps::cpuCompute(inputs.front(), outputs.front());
}
#ifdef HAVE_DNN_NGRAPH
virtual Ptr<BackendNode> initNgraph(const std::vector<Ptr<BackendWrapper>> &inputs,
const std::vector<Ptr<BackendNode>> &nodes) CV_OVERRIDE {
using namespace ov::op;
auto input_node = nodes[0].dynamicCast<InfEngineNgraphNode>()->node;
std::shared_ptr<ov::Node> output_node;
output_node = std::make_shared<v0::SpaceToDepth>(input_node, v0::SpaceToDepth::SpaceToDepthMode::BLOCKS_FIRST, static_cast<size_t>(blocksize));
return Ptr<BackendNode>(new InfEngineNgraphNode(output_node));
}
#endif // HAVE_DNN_NGRAPH
#ifdef HAVE_CUDA
Ptr<BackendNode> initCUDA(void *context_,
const std::vector<Ptr<BackendWrapper>> &inputs,
const std::vector<Ptr<BackendWrapper>> &outputs) override {
using namespace cv::dnn::cuda4dnn;
auto context = reinterpret_cast<csl::CSLContext*>(context_);
std::vector<size_t> perm(permutation.begin(), permutation.end());
return make_cuda_node<cuda4dnn::DepthSpaceOps>(preferableTarget, std::move(context->stream), internal_shape, perm);
}
#endif // HAVE_CUDA
#ifdef HAVE_CANN
virtual Ptr<BackendNode> initCann(const std::vector<Ptr<BackendWrapper>> &inputs,
const std::vector<Ptr<BackendWrapper>> &outputs,
const std::vector<Ptr<BackendNode>> &nodes) CV_OVERRIDE {
CV_CheckEQ(inputs.size(), static_cast<size_t>(1), "DepthToSpace/CANN: only accepts one input wrapper");
CV_CheckEQ(nodes.size(), static_cast<size_t>(1), "DepthToSpace/CANN: only accepts one input node");
auto input_tensor_wrapper = inputs.front().dynamicCast<CannBackendWrapper>();
auto input_tensor_desc = input_tensor_wrapper->getTensorDesc();
auto input_node = nodes.front().dynamicCast<CannBackendNode>()->getOp();
auto node = std::make_shared<ge::op::SpaceToDepth>(name);
node->set_attr_block_size(blocksize);
node->set_attr_data_format("NCHW");
node->set_input_x_by_name(*input_node, input_tensor_wrapper->name.c_str());
node->update_input_desc_x(*input_tensor_desc);
auto output_tensor_desc = std::make_shared<ge::TensorDesc>(ge::Shape(), ge::FORMAT_NCHW, ge::DT_FLOAT);
node->update_output_desc_y(*output_tensor_desc);
return Ptr<BackendNode>(new CannBackendNode(node));
}
#endif
#ifdef HAVE_TIMVX
virtual Ptr<BackendNode> initTimVX(void* timvx_info_,
const std::vector<Ptr<BackendWrapper>> &inputs,
const std::vector<Ptr<BackendWrapper>> &outputs,
bool isLast) CV_OVERRIDE {
auto info = reinterpret_cast<TimVXInfo*>(timvx_info_);
CV_Assert(info);
auto timvx_graph = info->getGraph();
CV_Assert(timvx_graph);
auto graph = timvx_graph->graph;
auto input_wrapper = inputs.front().dynamicCast<TimVXBackendWrapper>();
int input_wrapper_index = -1;
if (input_wrapper->isTensor()) {
input_wrapper_index = timvx_graph->getTensorIndex(input_wrapper->getTensor());
if (input_wrapper_index == -1) {
auto tmp = input_wrapper->getMat();
input_wrapper = std::make_shared<TimVXBackendWrapper>(tmp);
}
}
if (!input_wrapper->isTensor() || input_wrapper_index == 1) {
auto input_node_quant = Ptr<tim::vx::Quantization>(new tim::vx::Quantization(tim::vx::QuantType::ASYMMETRIC, 1.0f, 0));
input_wrapper->createTensor(graph, tim::vx::TensorAttribute::INPUT, input_node_quant);
input_wrapper_index = timvx_graph->addWrapper(input_wrapper);
}
auto output_wrapper = outputs.front().dynamicCast<TimVXBackendWrapper>();
auto output_node_quant = input_wrapper->getTensorQuantization();
if (isLast) {
auto shape_type = getShapeTypeFromMat(output_wrapper->getMat());
output_wrapper->setTensorShape(shape_type);
output_wrapper->createTensor(graph, tim::vx::TensorAttribute::OUTPUT, output_node_quant);
} else {
output_wrapper->createTensor(graph, tim::vx::TensorAttribute::TRANSIENT, output_node_quant);
}
int output_wrapper_index = timvx_graph->addWrapper(output_wrapper);
std::shared_ptr<tim::vx::Operation> timvx_node;
timvx_node = graph->CreateOperation<tim::vx::ops::SpaceToDepth>(std::vector<int>{blocksize, blocksize});
std::vector<int> input_wrapper_indices{input_wrapper_index}, output_wrapper_indices{output_wrapper_index};
return Ptr<BackendNode>(new TimVXBackendNode(timvx_graph, timvx_node, input_wrapper_indices, output_wrapper_indices));
}
#endif
private:
int blocksize;
};
Ptr<SpaceToDepthLayer> SpaceToDepthLayer::create(const LayerParams &params) {
return makePtr<SpaceToDepthLayerImpl>(params);
}
}} // namespace cv::dnn
modules/dnn/src/onnx/onnx_importer.cpp
+4 -78
View File
@@ -190,7 +190,7 @@ private:
void parseDetectionOutput (LayerParams& layerParams, const opencv_onnx::NodeProto& node_proto);
void parseCumSum (LayerParams& layerParams, const opencv_onnx::NodeProto& node_proto);
void parseElementWise (LayerParams& layerParams, const opencv_onnx::NodeProto& node_proto);
void parseDepthToSpace (LayerParams& layerParams, const opencv_onnx::NodeProto& node_proto);
void parseDepthSpaceOps (LayerParams& layerParams, const opencv_onnx::NodeProto& node_proto);
void parseRange (LayerParams& layerParams, const opencv_onnx::NodeProto& node_proto);
void parseScatter (LayerParams& layerParams, const opencv_onnx::NodeProto& node_proto);
void parseTile (LayerParams& layerParams, const opencv_onnx::NodeProto& node_proto);
@@ -2923,82 +2923,8 @@ void ONNXImporter::parseElementWise(LayerParams& layerParams, const opencv_onnx:
addLayer(layerParams, node_proto);
}
void ONNXImporter::parseDepthToSpace(LayerParams& layerParams, const opencv_onnx::NodeProto& node_proto_)
{
// We parse "DepthToSpace" and "SpaceToDepth" in this function.
opencv_onnx::NodeProto node_proto = node_proto_;
const std::string& layer_type = node_proto.op_type();
CV_Assert(layer_type == "DepthToSpace" || layer_type == "SpaceToDepth");
// Get blocksize
CV_Assert(layerParams.has("blocksize"));
int blocksize = layerParams.get<int>("blocksize");
CV_Assert(blocksize > 0);
// Get mode, only for "DepthToSpace"
std::string modeType = layerParams.get<std::string>("mode", "DCR");
MatShape inpShape = outShapes[node_proto.input(0)];
CV_Assert(inpShape.size() == 4);
int N = inpShape[0], C = inpShape[1], H = inpShape[2], W = inpShape[3];
// Implement DepthToSpace and SpaceToDepth by the Reshape and Permute layer.
std::array<int, 6> shape0, perm;
std::array<int, 4> shape1;
if (layer_type == "DepthToSpace")
{
if (modeType == "DCR")
{
shape0 = {N, blocksize, blocksize, C/(blocksize * blocksize), H, W};
perm = {0, 3, 4, 1, 5, 2};
shape1 = {N, C/(blocksize * blocksize), H * blocksize, W * blocksize};
}
else if (modeType == "CRD")
{
shape0 = {N, C/(blocksize * blocksize), blocksize, blocksize, H, W};
perm = {0, 1, 4, 2, 5, 3};
shape1 = {N, C/(blocksize * blocksize), H * blocksize, W * blocksize};
}
else
CV_Error(Error::StsNotImplemented, "The mode of " + modeType + " in " + layer_type + " Layer is not supported");
}
else // SpaceToDepth
{
shape0 = {N, C, H/blocksize, blocksize, W/blocksize, blocksize};
perm = {0, 3, 5, 1, 2, 4};
shape1 = {N, C * blocksize * blocksize, H/blocksize, W/blocksize};
}
// Step1: Reshape
LayerParams reshapeLp;
reshapeLp.name = layerParams.name + "/reshape";
reshapeLp.type = "Reshape";
CV_Assert(layer_id.find(reshapeLp.name) == layer_id.end());
reshapeLp.set("dim", DictValue::arrayInt(shape0.data(), shape0.size()));
opencv_onnx::NodeProto protoReshape;
protoReshape.add_input(node_proto.input(0));
protoReshape.add_output(reshapeLp.name);
addLayer(reshapeLp, protoReshape);
// Step2: Transpose
LayerParams permuteLp;
permuteLp.name = layerParams.name + "/permute";
permuteLp.type = "Permute";
CV_Assert(layer_id.find(permuteLp.name) == layer_id.end());
permuteLp.set("order", DictValue::arrayInt(perm.data(), perm.size()));
opencv_onnx::NodeProto protoPermute;
protoPermute.add_input(reshapeLp.name);
protoPermute.add_output(permuteLp.name);
addLayer(permuteLp, protoPermute);
// Step3: Reshape
layerParams.type = "Reshape";
layerParams.set("dim", DictValue::arrayInt(shape1.data(), shape1.size()));
node_proto.set_input(0, permuteLp.name);
void ONNXImporter::parseDepthSpaceOps(LayerParams &layerParams, const opencv_onnx::NodeProto& node_proto) {
CV_CheckTrue(layerParams.has("blocksize"), "blocksize is required but not found");
addLayer(layerParams, node_proto);
}
@@ -4035,7 +3961,7 @@ void ONNXImporter::buildDispatchMap_ONNX_AI(int opset_version)
dispatch["SoftMax"] = dispatch["Softmax"] = dispatch["LogSoftmax"] = &ONNXImporter::parseSoftMax;
dispatch["DetectionOutput"] = &ONNXImporter::parseDetectionOutput;
dispatch["CumSum"] = &ONNXImporter::parseCumSum;
dispatch["SpaceToDepth"] = dispatch["DepthToSpace"] = &ONNXImporter::parseDepthToSpace;
dispatch["SpaceToDepth"] = dispatch["DepthToSpace"] = &ONNXImporter::parseDepthSpaceOps;
dispatch["ScatterElements"] = dispatch["Scatter"] = dispatch["ScatterND"] = &ONNXImporter::parseScatter;
dispatch["Tile"] = &ONNXImporter::parseTile;
dispatch["LayerNormalization"] = &ONNXImporter::parseLayerNorm;
modules/dnn/test/test_int8_layers.cpp
+55
View File
@@ -517,6 +517,61 @@ TEST_P(Test_Int8_layers, Eltwise)
testLayer("split_max", "ONNX", 0.004, 0.012);
}
TEST_P(Test_Int8_layers, DepthSpaceOps) {
auto test_layer_with_onnx_conformance_models = [&](const std::string &model_name, double l1, double lInf) {
std::string model_path = _tf("onnx/conformance/node/test_" + model_name + "/model.onnx");
auto net = readNet(model_path);
// load reference inputs and outputs
std::string data_base_path = _tf("onnx/conformance/node/test_" + model_name + "/test_data_set_0");
Mat input = readTensorFromONNX(data_base_path + "/input_0.pb");
Mat ref_output = readTensorFromONNX(data_base_path + "/output_0.pb");
std::vector<float> input_scales, output_scales;
std::vector<int> input_zeropoints, output_zeropoints;
auto qnet = net.quantize(std::vector<Mat>{input}, CV_8S, CV_8S, false);
qnet.getInputDetails(input_scales, input_zeropoints);
qnet.getOutputDetails(output_scales, output_zeropoints);
qnet.setPreferableBackend(backend);
qnet.setPreferableTarget(target);
Mat quantized_input, quantized_output;
input.convertTo(quantized_input, CV_8S, 1.f / input_scales.front(), input_zeropoints.front());
qnet.setInput(quantized_input);
quantized_output = qnet.forward();
Mat output;
quantized_output.convertTo(output, CV_32F, output_scales.front(), -(output_scales.front() * output_zeropoints.front()));
normAssert(ref_output, output, model_name.c_str(), l1, lInf);
};
double l1 = default_l1, lInf = default_lInf;
{
l1 = 0.001; lInf = 0.002;
if (backend == DNN_BACKEND_TIMVX) { l1 = 0.001; lInf = 0.002; }
test_layer_with_onnx_conformance_models("spacetodepth", l1, lInf);
}
{
l1 = 0.022; lInf = 0.044;
if (backend == DNN_BACKEND_TIMVX) { l1 = 0.022; lInf = 0.044; }
test_layer_with_onnx_conformance_models("spacetodepth_example", l1, lInf);
}
{
l1 = 0.001; lInf = 0.002;
if (backend == DNN_BACKEND_TIMVX) { l1 = 0.24; lInf = 0.99; }
test_layer_with_onnx_conformance_models("depthtospace_crd_mode", l1, lInf);
}
test_layer_with_onnx_conformance_models("depthtospace_dcr_mode", 0.001, 0.002);
test_layer_with_onnx_conformance_models("depthtospace_example", 0.07, 0.14);
{
l1 = 0.07; lInf = 0.14;
if (backend == DNN_BACKEND_TIMVX) // diff too huge, l1 = 13.6; lInf = 27.2
applyTestTag(CV_TEST_TAG_DNN_SKIP_TIMVX);
test_layer_with_onnx_conformance_models("depthtospace_crd_mode_example", l1, lInf);
}
}
INSTANTIATE_TEST_CASE_P(/**/, Test_Int8_layers, dnnBackendsAndTargetsInt8());
class Test_Int8_nets : public DNNTestLayer
modules/dnn/test/test_onnx_conformance_layer_filter__openvino.inl.hpp
+10
View File
@@ -480,10 +480,20 @@ CASE(test_cumsum_2d_negative_axis)
// no filter
CASE(test_depthtospace_crd_mode)
// no filter
if (target == DNN_TARGET_OPENCL)
{
default_l1 = 1e-4; // Expected: (normL1) <= (l1), actual: 9.33057e-05 vs 1e-05
default_lInf = 2.5e-4; // Expected: (normInf) <= (lInf), actual: 0.000243843 vs 0.0001
}
CASE(test_depthtospace_crd_mode_example)
// no filter
CASE(test_depthtospace_dcr_mode)
// no filter
if (target == DNN_TARGET_OPENCL)
{
default_l1 = 1e-4; // Expected: (normL1) <= (l1), actual: 9.33057e-05 vs 1e-05
default_lInf = 2.5e-4; // Expected: (normInf) <= (lInf), actual: 0.000243843 vs 0.0001
}
CASE(test_depthtospace_example)
// no filter
CASE(test_dequantizelinear)
modules/gapi/include/opencv2/gapi/infer/bindings_onnx.hpp
+3
View File
@@ -51,6 +51,9 @@ public:
GAPI_WRAP
PyParams& cfgDisableMemPattern();
GAPI_WRAP
PyParams& cfgSessionOptions(const std::map<std::string, std::string>& options);
GBackend backend() const;
std::string tag() const;
cv::util::any params() const;
modules/gapi/include/opencv2/gapi/infer/onnx.hpp
+20 -1
View File
@@ -351,6 +351,7 @@ struct ParamDesc {
std::unordered_map<std::string, std::pair<cv::Scalar, cv::Scalar> > generic_mstd;
std::unordered_map<std::string, bool> generic_norm;
std::map<std::string, std::string> session_options;
std::vector<cv::gapi::onnx::ep::EP> execution_providers;
bool disable_mem_pattern;
};
@@ -634,6 +635,19 @@ public:
return *this;
}
/** @brief Configures session options for ONNX Runtime.
This function is used to set various session options for the ONNX Runtime
session by accepting a map of key-value pairs.
@param options A map of session option to be applied to the ONNX Runtime session.
@return the reference on modified object.
*/
Params<Net>& cfgSessionOptions(const std::map<std::string, std::string>& options) {
desc.session_options.insert(options.begin(), options.end());
return *this;
}
// BEGIN(G-API's network parametrization API)
GBackend backend() const { return cv::gapi::onnx::backend(); }
std::string tag() const { return Net::tag(); }
@@ -661,7 +675,7 @@ public:
@param model_path path to model file (.onnx file).
*/
Params(const std::string& tag, const std::string& model_path)
: desc{model_path, 0u, 0u, {}, {}, {}, {}, {}, {}, {}, {}, {}, true, {}, {}, {}, false }, m_tag(tag) {}
: desc{model_path, 0u, 0u, {}, {}, {}, {}, {}, {}, {}, {}, {}, true, {}, {}, {}, {}, false}, m_tag(tag) {}
/** @see onnx::Params::cfgMeanStdDev. */
void cfgMeanStdDev(const std::string &layer,
@@ -705,6 +719,11 @@ public:
desc.disable_mem_pattern = true;
}
/** @see onnx::Params::cfgSessionOptions. */
void cfgSessionOptions(const std::map<std::string, std::string>& options) {
desc.session_options.insert(options.begin(), options.end());
}
// BEGIN(G-API's network parametrization API)
GBackend backend() const { return cv::gapi::onnx::backend(); }
std::string tag() const { return m_tag; }
modules/gapi/src/backends/fluid/gfluidcore_func.simd.hpp
+1 -1
View File
@@ -2915,7 +2915,7 @@ CV_ALWAYS_INLINE
typename std::enable_if<DST_SHORT_OR_USHORT, void>::type
convertto_simd_nocoeff_impl(const uchar* inx, DST* outx)
{
v_uint8 a = vx_load(inx);
v_uint8 a = vx_load_low(inx);
v_uint16 res = v_expand_low(a);
store_i16(outx, res);
modules/gapi/src/backends/onnx/bindings_onnx.cpp
+6
View File
@@ -57,6 +57,12 @@ cv::gapi::onnx::PyParams::cfgDisableMemPattern() {
return *this;
}
cv::gapi::onnx::PyParams&
cv::gapi::onnx::PyParams::cfgSessionOptions(const std::map<std::string, std::string>& options) {
m_priv->cfgSessionOptions(options);
return *this;
}
cv::gapi::GBackend cv::gapi::onnx::PyParams::backend() const {
return m_priv->backend();
}
modules/gapi/src/backends/onnx/gonnxbackend.cpp
+4
View File
@@ -702,6 +702,10 @@ ONNXCompiled::ONNXCompiled(const gapi::onnx::detail::ParamDesc &pp)
cv::gimpl::onnx::addExecutionProvider(&session_options, ep);
}
for (const auto &option : pp.session_options) {
session_options.AddConfigEntry(option.first.c_str(), option.second.c_str());
}
if (pp.disable_mem_pattern) {
session_options.DisableMemPattern();
}
modules/highgui/CMakeLists.txt
+10
View File
@@ -49,6 +49,16 @@ list(REMOVE_ITEM highgui_ext_hdrs "${CMAKE_CURRENT_LIST_DIR}/include/opencv2/${n
set(OPENCV_HIGHGUI_BUILTIN_BACKEND "")
if(WITH_FRAMEBUFFER AND HAVE_FRAMEBUFFER)
set(OPENCV_HIGHGUI_BUILTIN_BACKEND "FB")
add_definitions(-DHAVE_FRAMEBUFFER)
list(APPEND highgui_srcs ${CMAKE_CURRENT_LIST_DIR}/src/window_framebuffer.cpp)
list(APPEND highgui_hdrs ${CMAKE_CURRENT_LIST_DIR}/src/window_framebuffer.hpp)
if(HAVE_FRAMEBUFFER_XVFB)
add_definitions(-DHAVE_FRAMEBUFFER_XVFB)
endif()
endif()
if(WITH_WAYLAND AND HAVE_WAYLAND)
set(OPENCV_HIGHGUI_BUILTIN_BACKEND "Wayland")
add_definitions(-DHAVE_WAYLAND)
modules/highgui/cmake/detect_framebuffer.cmake
+14
View File
@@ -0,0 +1,14 @@
# --- FB ---
set(HAVE_FRAMEBUFFER ON)
if(WITH_FRAMEBUFFER_XVFB)
try_compile(HAVE_FRAMEBUFFER_XVFB
"${CMAKE_CURRENT_BINARY_DIR}"
"${OpenCV_SOURCE_DIR}/cmake/checks/framebuffer.cpp")
if(HAVE_FRAMEBUFFER_XVFB)
message(STATUS "Check virtual framebuffer - done")
else()
message(STATUS
"Check virtual framebuffer - failed\n"
"Please install the xorg-x11-proto-devel or x11proto-dev package\n")
endif()
endif()
modules/highgui/cmake/init.cmake
+2
View File
@@ -39,6 +39,8 @@ endmacro()
add_backend("gtk" WITH_GTK)
add_backend("win32ui" WITH_WIN32UI)
add_backend("wayland" WITH_WAYLAND)
add_backend("framebuffer" WITH_FRAMEBUFFER)
# TODO cocoa
# TODO qt
# TODO opengl
modules/highgui/src/backend.hpp
+4
View File
@@ -127,6 +127,10 @@ std::shared_ptr<UIBackend> createUIBackendGTK();
std::shared_ptr<UIBackend> createUIBackendQT();
#endif
#ifdef HAVE_FRAMEBUFFER
std::shared_ptr<UIBackend> createUIBackendFramebuffer();
#endif
#endif // BUILD_PLUGIN
} // namespace highgui_backend
modules/highgui/src/registry.impl.hpp
+4
View File
@@ -44,6 +44,10 @@ std::vector<BackendInfo>& getBuiltinBackendsInfo()
DECLARE_DYNAMIC_BACKEND("GTK2")
#endif
#ifdef HAVE_FRAMEBUFFER
DECLARE_STATIC_BACKEND("FB", createUIBackendFramebuffer)
#endif
#if 0 // TODO
#ifdef HAVE_QT
DECLARE_STATIC_BACKEND("QT", createUIBackendQT)
modules/highgui/src/window_framebuffer.cpp
+798
View File
@@ -0,0 +1,798 @@
// This file is part of OpenCV project.
// It is subject to the license terms in the LICENSE file found in the top-level directory
// of this distribution and at http://opencv.org/license.html.
#include "precomp.hpp"
#include "window_framebuffer.hpp"
#include <opencv2/core/utils/configuration.private.hpp>
#include <opencv2/core/utils/logger.defines.hpp>
#ifdef NDEBUG
#define CV_LOG_STRIP_LEVEL CV_LOG_LEVEL_DEBUG + 1
#else
#define CV_LOG_STRIP_LEVEL CV_LOG_LEVEL_VERBOSE + 1
#endif
#include <opencv2/core/utils/logger.hpp>
#include <unistd.h>
#include <stdio.h>
#include <termios.h>
#include <fcntl.h>
#include <stdlib.h>
#include <linux/fb.h>
#include <linux/input.h>
#include <sys/mman.h>
#include <sys/ioctl.h>
#include "opencv2/imgproc.hpp"
#ifdef HAVE_FRAMEBUFFER_XVFB
#include <X11/XWDFile.h>
#include <X11/X.h>
#define C32INT(ptr) ((((unsigned char*)ptr)[0] << 24) | (((unsigned char*)ptr)[1] << 16) | \
(((unsigned char*)ptr)[2] << 8) | (((unsigned char*)ptr)[3] << 0))
#endif
namespace cv {
namespace highgui_backend {
std::shared_ptr<UIBackend> createUIBackendFramebuffer()
{
return std::make_shared<FramebufferBackend>();
}
static std::string& getFBMode()
{
static std::string fbModeOpenCV =
cv::utils::getConfigurationParameterString("OPENCV_HIGHGUI_FB_MODE", "FB");
return fbModeOpenCV;
}
static std::string& getFBFileName()
{
static std::string fbFileNameFB =
cv::utils::getConfigurationParameterString("FRAMEBUFFER", "/dev/fb0");
static std::string fbFileNameOpenCV =
cv::utils::getConfigurationParameterString("OPENCV_HIGHGUI_FB_DEVICE", "");
if (!fbFileNameOpenCV.empty()) return fbFileNameOpenCV;
return fbFileNameFB;
}
FramebufferWindow::FramebufferWindow(FramebufferBackend &_backend, int _flags):
backend(_backend), flags(_flags)
{
CV_LOG_DEBUG(NULL, "UI: FramebufferWindow::FramebufferWindow()");
FB_ID = "FramebufferWindow";
windowRect = Rect(0,0, backend.getFBWidth(), backend.getFBHeight());
}
FramebufferWindow::~FramebufferWindow()
{
CV_LOG_DEBUG(NULL, "UI: FramebufferWindow::~FramebufferWindow()");
}
void FramebufferWindow::imshow(InputArray image)
{
CV_LOG_DEBUG(NULL, "UI: FramebufferWindow::imshow(InputArray image)");
currentImg = image.getMat().clone();
CV_LOG_INFO(NULL, "UI: InputArray image: "
<< cv::typeToString(image.type()) << " size " << image.size());
if (currentImg.empty())
{
CV_LOG_WARNING(NULL, "UI: image is empty");
return;
}
CV_CheckEQ(currentImg.dims, 2, "UI: dims != 2");
Mat img = image.getMat();
switch (img.channels())
{
case 1:
{
Mat tmp;
switch(img.type())
{
case CV_8U:
tmp = img;
break;
case CV_8S:
cv::convertScaleAbs(img, tmp, 1, 127);
break;
case CV_16S:
cv::convertScaleAbs(img, tmp, 1/255., 127);
break;
case CV_16U:
cv::convertScaleAbs(img, tmp, 1/255.);
break;
case CV_32F:
case CV_64F: // assuming image has values in range [0, 1)
img.convertTo(tmp, CV_8U, 255., 0.);
break;
}
Mat rgb(img.rows, img.cols, CV_8UC3);
cvtColor(tmp, rgb, COLOR_GRAY2RGB);
img = rgb;
}
break;
case 3:
case 4:
{
Mat tmp(img.rows, img.cols, CV_8UC3);
convertToShow(img, tmp, true);
img = tmp;
}
break;
default:
CV_Error(cv::Error::StsBadArg, "Bad image: wrong number of channels");
}
{
Mat bgra(img.rows, img.cols, CV_8UC4);
cvtColor(img, bgra, COLOR_RGB2BGRA, bgra.channels());
img = bgra;
}
int newWidth = windowRect.width;
int newHeight = windowRect.height;
int cntChannel = img.channels();
cv::Size imgSize = currentImg.size();
if (flags & WINDOW_AUTOSIZE)
{
windowRect.width = imgSize.width;
windowRect.height = imgSize.height;
newWidth = windowRect.width;
newHeight = windowRect.height;
}
if (flags & WINDOW_FREERATIO)
{
newWidth = windowRect.width;
newHeight = windowRect.height;
}
else //WINDOW_KEEPRATIO
{
double aspect_ratio = ((double)img.cols) / img.rows;
newWidth = windowRect.width;
newHeight = (int)(windowRect.width / aspect_ratio);
if (newHeight > windowRect.height)
{
newWidth = (int)(windowRect.height * aspect_ratio);
newHeight = windowRect.height;
}
}
if ((newWidth != img.cols) && (newHeight != img.rows))
{
Mat imResize;
cv::resize(img, imResize, cv::Size(newWidth, newHeight), INTER_LINEAR);
img = imResize;
}
CV_LOG_INFO(NULL, "UI: Formated image: "
<< cv::typeToString(img.type()) << " size " << img.size());
if (backend.getMode() == FB_MODE_EMU)
{
CV_LOG_WARNING(NULL, "UI: FramebufferWindow::imshow is used in EMU mode");
return;
}
if (backend.getFBPointer() == MAP_FAILED)
{
CV_LOG_ERROR(NULL, "UI: Framebuffer is not mapped");
return;
}
int xOffset = backend.getFBXOffset();
int yOffset = backend.getFBYOffset();
int fbHeight = backend.getFBHeight();
int fbWidth = backend.getFBWidth();
int lineLength = backend.getFBLineLength();
int img_start_x;
int img_start_y;
int img_end_x;
int img_end_y;
int fb_start_x;
int fb_start_y;
if (windowRect.y - yOffset < 0)
{
img_start_y = - (windowRect.y - yOffset);
}
else
{
img_start_y = 0;
}
if (windowRect.x - xOffset < 0)
{
img_start_x = - (windowRect.x - xOffset);
}
else
{
img_start_x = 0;
}
if (windowRect.y + yOffset + img.rows > fbHeight)
{
img_end_y = fbHeight - windowRect.y - yOffset;
}
else
{
img_end_y = img.rows;
}
if (windowRect.x + xOffset + img.cols > fbWidth)
{
img_end_x = fbWidth - windowRect.x - xOffset;
}
else
{
img_end_x = img.cols;
}
if (windowRect.y + yOffset >= 0)
{
fb_start_y = windowRect.y + yOffset;
}
else
{
fb_start_y = 0;
}
if (windowRect.x + xOffset >= 0)
{
fb_start_x = windowRect.x + xOffset;
}
else
{
fb_start_x = 0;
}
for (int y = img_start_y; y < img_end_y; y++)
{
std::memcpy(backend.getFBPointer() +
(fb_start_y + y - img_start_y) * lineLength + fb_start_x * cntChannel,
img.ptr<unsigned char>(y) + img_start_x * cntChannel,
(img_end_x - img_start_x) * cntChannel);
}
}
double FramebufferWindow::getProperty(int /*prop*/) const
{
CV_LOG_WARNING(NULL, "UI: getProperty (not supported)");
return 0.0;
}
bool FramebufferWindow::setProperty(int /*prop*/, double /*value*/)
{
CV_LOG_WARNING(NULL, "UI: setProperty (not supported)");
return false;
}
void FramebufferWindow::resize(int width, int height)
{
CV_LOG_DEBUG(NULL, "UI: FramebufferWindow::resize(int width "
<< width <<", height " << height << ")");
CV_Assert(width > 0);
CV_Assert(height > 0);
if (!(flags & WINDOW_AUTOSIZE))
{
windowRect.width = width;
windowRect.height = height;
if (!currentImg.empty())
{
imshow(currentImg);
}
}
}
void FramebufferWindow::move(int x, int y)
{
CV_LOG_DEBUG(NULL, "UI: FramebufferWindow::move(int x " << x << ", y " << y <<")");
windowRect.x = x;
windowRect.y = y;
if (!currentImg.empty())
{
imshow(currentImg);
}
}
Rect FramebufferWindow::getImageRect() const
{
CV_LOG_DEBUG(NULL, "UI: FramebufferWindow::getImageRect()");
return windowRect;
}
void FramebufferWindow::setTitle(const std::string& /*title*/)
{
CV_LOG_WARNING(NULL, "UI: setTitle (not supported)");
}
void FramebufferWindow::setMouseCallback(MouseCallback /*onMouse*/, void* /*userdata*/)
{
CV_LOG_WARNING(NULL, "UI: setMouseCallback (not supported)");
}
std::shared_ptr<UITrackbar> FramebufferWindow::createTrackbar(
const std::string& /*name*/,
int /*count*/,
TrackbarCallback /*onChange*/,
void* /*userdata*/)
{
CV_LOG_WARNING(NULL, "UI: createTrackbar (not supported)");
return nullptr;
}
std::shared_ptr<UITrackbar> FramebufferWindow::findTrackbar(const std::string& /*name*/)
{
CV_LOG_WARNING(NULL, "UI: findTrackbar (not supported)");
return nullptr;
}
const std::string& FramebufferWindow::getID() const
{
CV_LOG_DEBUG(NULL, "UI: FramebufferWindow::getID()");
return FB_ID;
}
bool FramebufferWindow::isActive() const
{
CV_LOG_DEBUG(NULL, "UI: FramebufferWindow::isActive()");
return true;
}
void FramebufferWindow::destroy()
{
CV_LOG_DEBUG(NULL, "UI: FramebufferWindow::destroy()");
}
int FramebufferBackend::fbOpenAndGetInfo()
{
std::string fbFileName = getFBFileName();
CV_LOG_INFO(NULL, "UI: FramebufferWindow::The following is used as a framebuffer file: \n"
<< fbFileName);
int fb_fd = open(fbFileName.c_str(), O_RDWR);
if (fb_fd == -1)
{
CV_LOG_ERROR(NULL, "UI: can't open framebuffer");
return -1;
}
if (ioctl(fb_fd, FBIOGET_FSCREENINFO, &fixInfo))
{
CV_LOG_ERROR(NULL, "UI: can't read fix info for framebuffer");
return -1;
}
if (ioctl(fb_fd, FBIOGET_VSCREENINFO, &varInfo))
{
CV_LOG_ERROR(NULL, "UI: can't read var info for framebuffer");
return -1;
}
CV_LOG_INFO(NULL, "UI: framebuffer info: \n"
<< " red offset " << varInfo.red.offset << " length " << varInfo.red.length << "\n"
<< " green offset " << varInfo.green.offset << " length " << varInfo.green.length << "\n"
<< " blue offset " << varInfo.blue.offset << " length " << varInfo.blue.length << "\n"
<< "transp offset " << varInfo.transp.offset << " length " <<varInfo.transp.length << "\n"
<< "bits_per_pixel " << varInfo.bits_per_pixel);
if ((varInfo.red.offset != 16) && (varInfo.red.length != 8) &&
(varInfo.green.offset != 8) && (varInfo.green.length != 8) &&
(varInfo.blue.offset != 0) && (varInfo.blue.length != 8) &&
(varInfo.bits_per_pixel != 32) )
{
close(fb_fd);
CV_LOG_ERROR(NULL, "UI: Framebuffer format is not supported "
<< "(use BGRA format with bits_per_pixel = 32)");
return -1;
}
fbWidth = varInfo.xres;
fbHeight = varInfo.yres;
fbXOffset = varInfo.xoffset;
fbYOffset = varInfo.yoffset;
fbBitsPerPixel = varInfo.bits_per_pixel;
fbLineLength = fixInfo.line_length;
fbScreenSize = max(varInfo.xres, varInfo.xres_virtual) *
max(varInfo.yres, varInfo.yres_virtual) *
fbBitsPerPixel / 8;
fbPointer = (unsigned char*)
mmap(0, fbScreenSize, PROT_READ | PROT_WRITE, MAP_SHARED, fb_fd, 0);
if (fbPointer == MAP_FAILED)
{
CV_LOG_ERROR(NULL, "UI: can't mmap framebuffer");
return -1;
}
return fb_fd;
}
int FramebufferBackend::XvfbOpenAndGetInfo()
{
int fb_fd = -1;
#ifdef HAVE_FRAMEBUFFER_XVFB
std::string fbFileName = getFBFileName();
CV_LOG_INFO(NULL, "UI: FramebufferWindow::The following is used as a framebuffer file: \n"
<< fbFileName);
fb_fd = open(fbFileName.c_str(), O_RDWR);
if (fb_fd == -1)
{
CV_LOG_ERROR(NULL, "UI: can't open framebuffer");
return -1;
}
XWDFileHeader *xwd_header;
xwd_header = (XWDFileHeader*)
mmap(NULL, sizeof(XWDFileHeader), PROT_READ, MAP_SHARED, fb_fd, 0);
if (xwd_header == MAP_FAILED)
{
CV_LOG_ERROR(NULL, "UI: can't mmap xwd header");
return -1;
}
if (C32INT(&(xwd_header->pixmap_format)) != ZPixmap)
{
CV_LOG_ERROR(NULL, "Unsupported pixmap format: " << xwd_header->pixmap_format);
return -1;
}
if (xwd_header->xoffset != 0)
{
CV_LOG_ERROR(NULL, "UI: Unsupported xoffset value: " << xwd_header->xoffset );
return -1;
}
unsigned int r = C32INT(&(xwd_header->red_mask));
unsigned int g = C32INT(&(xwd_header->green_mask));
unsigned int b = C32INT(&(xwd_header->blue_mask));
fbWidth = C32INT(&(xwd_header->pixmap_width));
fbHeight = C32INT(&(xwd_header->pixmap_height));
fbXOffset = 0;
fbYOffset = 0;
fbLineLength = C32INT(&(xwd_header->bytes_per_line));
fbBitsPerPixel = C32INT(&(xwd_header->bits_per_pixel));
CV_LOG_INFO(NULL, "UI: XVFB info: \n"
<< " red_mask " << r << "\n"
<< " green_mask " << g << "\n"
<< " blue_mask " << b << "\n"
<< "bits_per_pixel " << fbBitsPerPixel);
if ((r != 16711680 ) && (g != 65280 ) && (b != 255 ) &&
(fbBitsPerPixel != 32))
{
CV_LOG_ERROR(NULL, "UI: Framebuffer format is not supported "
<< "(use BGRA format with bits_per_pixel = 32)");
return -1;
}
xvfb_len_header = C32INT(&(xwd_header->header_size));
xvfb_len_colors = sizeof(XWDColor) * C32INT(&(xwd_header->ncolors));
xvfb_len_pixmap = C32INT(&(xwd_header->bytes_per_line)) *
C32INT(&(xwd_header->pixmap_height));
munmap(xwd_header, sizeof(XWDFileHeader));
fbScreenSize = xvfb_len_header + xvfb_len_colors + xvfb_len_pixmap;
xwd_header = (XWDFileHeader*)
mmap(NULL, fbScreenSize, PROT_READ | PROT_WRITE, MAP_SHARED, fb_fd, 0);
fbPointer = (unsigned char*)xwd_header;
fbPointer_dist = xvfb_len_header + xvfb_len_colors;
#else
CV_LOG_WARNING(NULL, "UI: To use virtual framebuffer, "
<< "compile OpenCV with the WITH_FRAMEBUFFER_XVFB=ON");
#endif
return fb_fd;
}
fb_var_screeninfo &FramebufferBackend::getVarInfo()
{
return varInfo;
}
fb_fix_screeninfo &FramebufferBackend::getFixInfo()
{
return fixInfo;
}
int FramebufferBackend::getFramebuffrerID()
{
return fbID;
}
int FramebufferBackend::getFBWidth()
{
return fbWidth;
}
int FramebufferBackend::getFBHeight()
{
return fbHeight;
}
int FramebufferBackend::getFBXOffset()
{
return fbXOffset;
}
int FramebufferBackend::getFBYOffset()
{
return fbYOffset;
}
int FramebufferBackend::getFBBitsPerPixel()
{
return fbBitsPerPixel;
}
int FramebufferBackend::getFBLineLength()
{
return fbLineLength;
}
unsigned char* FramebufferBackend::getFBPointer()
{
return fbPointer + fbPointer_dist;
}
Mat& FramebufferBackend::getBackgroundBuff()
{
return backgroundBuff;
}
OpenCVFBMode FramebufferBackend::getMode()
{
return mode;
}
FramebufferBackend::FramebufferBackend():mode(FB_MODE_FB), fbPointer_dist(0)
{
CV_LOG_DEBUG(NULL, "UI: FramebufferWindow::FramebufferBackend()");
std::string fbModeStr = getFBMode();
if (fbModeStr == "EMU")
{
mode = FB_MODE_EMU;
CV_LOG_WARNING(NULL, "UI: FramebufferWindow is trying to use EMU mode");
}
if (fbModeStr == "FB")
{
mode = FB_MODE_FB;
CV_LOG_WARNING(NULL, "UI: FramebufferWindow is trying to use FB mode");
}
if (fbModeStr == "XVFB")
{
mode = FB_MODE_XVFB;
CV_LOG_WARNING(NULL, "UI: FramebufferWindow is trying to use XVFB mode");
}
fbID = -1;
if (mode == FB_MODE_FB)
{
fbID = fbOpenAndGetInfo();
}
if (mode == FB_MODE_XVFB)
{
fbID = XvfbOpenAndGetInfo();
}
CV_LOG_INFO(NULL, "UI: FramebufferWindow::fbID " << fbID);
if (fbID == -1)
{
mode = FB_MODE_EMU;
fbWidth = 640;
fbHeight = 480;
fbXOffset = 0;
fbYOffset = 0;
fbBitsPerPixel = 0;
fbLineLength = 0;
CV_LOG_WARNING(NULL, "UI: FramebufferWindow is used in EMU mode");
return;
}
CV_LOG_INFO(NULL, "UI: Framebuffer's width, height, bits per pix: "
<< fbWidth << " " << fbHeight << " " << fbBitsPerPixel);
CV_LOG_INFO(NULL, "UI: Framebuffer's offsets (x, y), line length: "
<< fbXOffset << " " << fbYOffset << " " << fbLineLength);
backgroundBuff = Mat(fbHeight, fbWidth, CV_8UC4);
int cntChannel = 4;
for (int y = fbYOffset; y < backgroundBuff.rows + fbYOffset; y++)
{
std::memcpy(backgroundBuff.ptr<unsigned char>(y - fbYOffset),
getFBPointer() + y * fbLineLength + fbXOffset * cntChannel,
backgroundBuff.cols * cntChannel);
}
}
FramebufferBackend::~FramebufferBackend()
{
CV_LOG_DEBUG(NULL, "UI: FramebufferBackend::~FramebufferBackend()");
if(fbID == -1) return;
if (fbPointer != MAP_FAILED)
{
int cntChannel = 4;
for (int y = fbYOffset; y < backgroundBuff.rows + fbYOffset; y++)
{
std::memcpy(getFBPointer() + y * fbLineLength + fbXOffset * cntChannel,
backgroundBuff.ptr<cv::Vec4b>(y - fbYOffset),
backgroundBuff.cols * cntChannel);
}
munmap(fbPointer, fbScreenSize);
}
close(fbID);
}
void FramebufferBackend::destroyAllWindows() {
CV_LOG_DEBUG(NULL, "UI: FramebufferBackend::destroyAllWindows()");
}
// namedWindow
std::shared_ptr<UIWindow> FramebufferBackend::createWindow(
const std::string& winname,
int flags)
{
CV_LOG_DEBUG(NULL, "UI: FramebufferBackend::createWindow("
<< winname << ", " << flags << ")");
return std::make_shared<FramebufferWindow>(*this, flags);
}
void FramebufferBackend::initTermios(int echo, int wait)
{
tcgetattr(0, &old);
current = old;
current.c_lflag &= ~ICANON;
current.c_lflag &= ~ISIG;
current.c_cc[VMIN] = wait;
if (echo)
{
current.c_lflag |= ECHO;
}
else
{
current.c_lflag &= ~ECHO;
}
tcsetattr(0, TCSANOW, &current);
}
void FramebufferBackend::resetTermios(void)
{
tcsetattr(0, TCSANOW, &old);
}
int FramebufferBackend::getch_(int echo, int wait)
{
int ch;
initTermios(echo, wait);
ch = getchar();
if (ch < 0)
{
rewind(stdin);
}
resetTermios();
return ch;
}
bool FramebufferBackend::kbhit()
{
int byteswaiting = 0;
initTermios(0, 1);
if (ioctl(0, FIONREAD, &byteswaiting) < 0)
{
CV_LOG_ERROR(NULL, "UI: Framebuffer ERR byteswaiting" );
}
resetTermios();
return byteswaiting > 0;
}
int FramebufferBackend::waitKeyEx(int delay)
{
CV_LOG_DEBUG(NULL, "UI: FramebufferBackend::waitKeyEx(int delay = " << delay << ")");
int code = -1;
if (delay <= 0)
{
int ch = getch_(0, 1);
CV_LOG_INFO(NULL, "UI: FramebufferBackend::getch_() take value = " << (int)ch);
code = ch;
while ((ch = getch_(0, 0)) >= 0)
{
CV_LOG_INFO(NULL, "UI: FramebufferBackend::getch_() take value = "
<< (int)ch << " (additional code on <stdin>)");
code = ch;
}
}
else
{
bool f_kbhit = false;
while (!(f_kbhit = kbhit()) && (delay > 0))
{
delay -= 1;
usleep(1000);
}
if (f_kbhit)
{
CV_LOG_INFO(NULL, "UI: FramebufferBackend kbhit is True ");
int ch = getch_(0, 1);
CV_LOG_INFO(NULL, "UI: FramebufferBackend::getch_() take value = " << (int)ch);
code = ch;
while ((ch = getch_(0, 0)) >= 0)
{
CV_LOG_INFO(NULL, "UI: FramebufferBackend::getch_() take value = "
<< (int)ch << " (additional code on <stdin>)");
code = ch;
}
}
}
CV_LOG_INFO(NULL, "UI: FramebufferBackend::waitKeyEx() result code = " << code);
return code;
}
int FramebufferBackend::pollKey()
{
CV_LOG_DEBUG(NULL, "UI: FramebufferBackend::pollKey()");
int code = -1;
bool f_kbhit = false;
f_kbhit = kbhit();
if (f_kbhit)
{
CV_LOG_INFO(NULL, "UI: FramebufferBackend kbhit is True ");
int ch = getch_(0, 1);
CV_LOG_INFO(NULL, "UI: FramebufferBackend::getch_() take value = " << (int)ch);
code = ch;
while ((ch = getch_(0, 0)) >= 0)
{
CV_LOG_INFO(NULL, "UI: FramebufferBackend::getch_() take value = "
<< (int)ch << " (additional code on <stdin>)");
code = ch;
}
}
return code;
}
const std::string FramebufferBackend::getName() const
{
return "FB";
}
}} // cv::highgui_backend::
modules/highgui/src/window_framebuffer.hpp
+135
View File
@@ -0,0 +1,135 @@
// This file is part of OpenCV project.
// It is subject to the license terms in the LICENSE file found in the top-level directory
// of this distribution and at http://opencv.org/license.html.
#ifndef OPENCV_HIGHGUI_WINDOWS_FRAMEBUFFER_HPP
#define OPENCV_HIGHGUI_WINDOWS_FRAMEBUFFER_HPP
#include "backend.hpp"
#include <linux/fb.h>
#include <linux/input.h>
#include <termios.h>
namespace cv {
namespace highgui_backend {
enum OpenCVFBMode{
FB_MODE_EMU,
FB_MODE_FB,
FB_MODE_XVFB
};
class FramebufferBackend;
class FramebufferWindow : public UIWindow
{
FramebufferBackend &backend;
std::string FB_ID;
Rect windowRect;
int flags;
Mat currentImg;
public:
FramebufferWindow(FramebufferBackend &backend, int flags);
virtual ~FramebufferWindow();
virtual void imshow(InputArray image) override;
virtual double getProperty(int prop) const override;
virtual bool setProperty(int prop, double value) override;
virtual void resize(int width, int height) override;
virtual void move(int x, int y) override;
virtual Rect getImageRect() const override;
virtual void setTitle(const std::string& title) override;
virtual void setMouseCallback(MouseCallback onMouse, void* userdata /*= 0*/) override;
virtual std::shared_ptr<UITrackbar> createTrackbar(
const std::string& name,
int count,
TrackbarCallback onChange /*= 0*/,
void* userdata /*= 0*/
) override;
virtual std::shared_ptr<UITrackbar> findTrackbar(const std::string& name) override;
virtual const std::string& getID() const override;
virtual bool isActive() const override;
virtual void destroy() override;
}; // FramebufferWindow
class FramebufferBackend: public UIBackend
{
OpenCVFBMode mode;
struct termios old, current;
void initTermios(int echo, int wait);
void resetTermios(void);
int getch_(int echo, int wait);
bool kbhit();
fb_var_screeninfo varInfo;
fb_fix_screeninfo fixInfo;
int fbWidth;
int fbHeight;
int fbXOffset;
int fbYOffset;
int fbBitsPerPixel;
int fbLineLength;
long int fbScreenSize;
unsigned char* fbPointer;
unsigned int fbPointer_dist;
Mat backgroundBuff;
int fbOpenAndGetInfo();
int fbID;
unsigned int xvfb_len_header;
unsigned int xvfb_len_colors;
unsigned int xvfb_len_pixmap;
int XvfbOpenAndGetInfo();
public:
fb_var_screeninfo &getVarInfo();
fb_fix_screeninfo &getFixInfo();
int getFramebuffrerID();
int getFBWidth();
int getFBHeight();
int getFBXOffset();
int getFBYOffset();
int getFBBitsPerPixel();
int getFBLineLength();
unsigned char* getFBPointer();
Mat& getBackgroundBuff();
OpenCVFBMode getMode();
FramebufferBackend();
virtual ~FramebufferBackend();
virtual void destroyAllWindows()override;
// namedWindow
virtual std::shared_ptr<UIWindow> createWindow(
const std::string& winname,
int flags
)override;
virtual int waitKeyEx(int delay /*= 0*/)override;
virtual int pollKey() override;
virtual const std::string getName() const override;
};
}} // cv::highgui_backend::
#endif
modules/highgui/src/window_w32.cpp
+1 -1
View File
@@ -676,7 +676,7 @@ void cvSetPropTopmost_W32(const char* name, const bool topmost)
static bool setPropTopmost_(CvWindow& window, bool topmost)
{
HWND flag = topmost ? HWND_TOPMOST : HWND_TOP;
HWND flag = topmost ? HWND_TOPMOST : HWND_NOTOPMOST;
BOOL success = SetWindowPos(window.frame, flag, 0, 0, 0, 0, SWP_NOMOVE | SWP_NOSIZE);
if (!success)
modules/highgui/test/test_gui.cpp
+4 -2
View File
@@ -148,7 +148,8 @@ static void Foo(int, void* counter)
&& !defined HAVE_WIN32UI \
&& !defined HAVE_WAYLAND \
) \
|| defined(__APPLE__) // test fails on Mac (cocoa)
|| defined(__APPLE__) /* test fails on Mac (cocoa) */ \
|| defined HAVE_FRAMEBUFFER /* trackbar is not supported */
TEST(Highgui_GUI, DISABLED_trackbar_unsafe)
#else
TEST(Highgui_GUI, trackbar_unsafe)
@@ -188,7 +189,8 @@ void testTrackbarCallback(int pos, void* param)
&& !defined HAVE_WIN32UI \
&& !defined HAVE_WAYLAND \
) \
|| defined(__APPLE__) // test fails on Mac (cocoa)
|| defined(__APPLE__) /* test fails on Mac (cocoa) */ \
|| defined HAVE_FRAMEBUFFER /* trackbar is not supported */
TEST(Highgui_GUI, DISABLED_trackbar)
#else
TEST(Highgui_GUI, trackbar)
modules/imgcodecs/src/grfmt_hdr.cpp
+3
View File
@@ -59,6 +59,9 @@ HdrDecoder::HdrDecoder()
HdrDecoder::~HdrDecoder()
{
if(file) {
fclose(file);
}
}
size_t HdrDecoder::signatureLength() const
modules/imgproc/include/opencv2/imgproc.hpp
+1 -1
View File
@@ -4269,7 +4269,7 @@ Examples of how intersectConvexConvex works
When false, no intersection is found. If the polygons share a side or the vertex of one polygon lies on an edge
of the other, they are not considered nested and an intersection will be found regardless of the value of handleNested.
@returns Absolute value of area of intersecting polygon
@returns Area of intersecting polygon. May be negative, if algorithm has not converged, e.g. non-convex input.
@note intersectConvexConvex doesn't confirm that both polygons are convex and will return invalid results if they aren't.
*/
modules/imgproc/perf/perf_bilateral.cpp
+5 -3
View File
@@ -7,23 +7,25 @@ namespace opencv_test {
CV_ENUM(Mat_Type, CV_8UC1, CV_8UC3, CV_32FC1, CV_32FC3)
typedef TestBaseWithParam< tuple<Size, int, Mat_Type> > TestBilateralFilter;
typedef TestBaseWithParam< tuple<Size, int, Mat_Type, double> > TestBilateralFilter;
PERF_TEST_P( TestBilateralFilter, BilateralFilter,
Combine(
Values( szVGA, sz1080p ), // image size
Values( 3, 5 ), // d
Mat_Type::all() // image type
Mat_Type::all(), // image type
Values(1., 5.)
)
)
{
Size sz;
int d, type;
const double sigmaColor = 1., sigmaSpace = 1.;
double sigmaColor, sigmaSpace;
sz = get<0>(GetParam());
d = get<1>(GetParam());
type = get<2>(GetParam());
sigmaColor = sigmaSpace = get<3>(GetParam());
Mat src(sz, type);
Mat dst(sz, type);
modules/imgproc/src/color_hsv.simd.hpp
+2 -2
View File
@@ -130,7 +130,7 @@ struct RGB2HSV_b
// sdiv = sdiv_table[v]
v_int32 sdiv0, sdiv1, sdiv2, sdiv3;;
v_uint16 vd0, vd1, vd2;
v_uint16 vd0, vd1;
v_expand(v, vd0, vd1);
v_int32 vq0, vq1, vq2, vq3;
v_expand(v_reinterpret_as_s16(vd0), vq0, vq1);
@@ -150,7 +150,7 @@ struct RGB2HSV_b
// hdiv = hdiv_table[diff]
v_int32 hdiv0, hdiv1, hdiv2, hdiv3;
v_uint16 diffd0, diffd1, diffd2;
v_uint16 diffd0, diffd1;
v_expand(diff, diffd0, diffd1);
v_int32 diffq0, diffq1, diffq2, diffq3;
v_expand(v_reinterpret_as_s16(diffd0), diffq0, diffq1);
modules/imgproc/src/demosaicing.cpp
+3 -9
View File
@@ -593,10 +593,10 @@ class Bayer2Gray_Invoker :
public ParallelLoopBody
{
public:
Bayer2Gray_Invoker(const Mat& _srcmat, Mat& _dstmat, int _start_with_green, bool _brow,
Bayer2Gray_Invoker(const Mat& _srcmat, Mat& _dstmat, int _start_with_green,
const Size& _size, int _bcoeff, int _rcoeff) :
ParallelLoopBody(), srcmat(_srcmat), dstmat(_dstmat), Start_with_green(_start_with_green),
Brow(_brow), size(_size), Bcoeff(_bcoeff), Rcoeff(_rcoeff)
size(_size), Bcoeff(_bcoeff), Rcoeff(_rcoeff)
{
}
@@ -612,13 +612,11 @@ public:
int dst_step = (int)(dstmat.step/sizeof(T));
int bcoeff = Bcoeff, rcoeff = Rcoeff;
int start_with_green = Start_with_green;
bool brow = Brow;
dst0 += dst_step + 1;
if (range.start % 2)
{
brow = !brow;
std::swap(bcoeff, rcoeff);
start_with_green = !start_with_green;
}
@@ -680,7 +678,6 @@ public:
dst0[-1] = dst0[0];
dst0[size.width] = dst0[size.width-1];
brow = !brow;
std::swap(bcoeff, rcoeff);
start_with_green = !start_with_green;
}
@@ -690,7 +687,6 @@ private:
Mat srcmat;
Mat dstmat;
int Start_with_green;
bool Brow;
Size size;
int Bcoeff, Rcoeff;
};
@@ -704,11 +700,9 @@ static void Bayer2Gray_( const Mat& srcmat, Mat& dstmat, int code )
Size size = srcmat.size();
int bcoeff = B2Y, rcoeff = R2Y;
int start_with_green = code == COLOR_BayerGB2GRAY || code == COLOR_BayerGR2GRAY;
bool brow = true;
if( code != COLOR_BayerBG2GRAY && code != COLOR_BayerGB2GRAY )
{
brow = false;
std::swap(bcoeff, rcoeff);
}
size.height -= 2;
@@ -718,7 +712,7 @@ static void Bayer2Gray_( const Mat& srcmat, Mat& dstmat, int code )
{
Range range(0, size.height);
Bayer2Gray_Invoker<T, SIMDInterpolator> invoker(srcmat, dstmat,
start_with_green, brow, size, bcoeff, rcoeff);
start_with_green, size, bcoeff, rcoeff);
parallel_for_(range, invoker, dstmat.total()/static_cast<double>(1<<16));
}
modules/imgproc/src/geometry.cpp
+27 -8
View File
@@ -318,9 +318,12 @@ static void addSharedSeg( Point2f p, Point2f q, Point2f*& result )
*result++ = q;
}
// Note: The function and subroutings use direct pointer arithmetics instead of arrays indexing.
// Each loop iteration may push to result array up to 3 times.
// It means that we need +3 spare result elements against result_size
// to catch agorithmic overflow and prevent actual result array overflow.
static int intersectConvexConvex_( const Point2f* P, int n, const Point2f* Q, int m,
Point2f* result, float* _area )
Point2f* result, int result_size, float* _area )
{
Point2f* result0 = result;
// P has n vertices, Q has m vertices.
@@ -398,7 +401,7 @@ static int intersectConvexConvex_( const Point2f* P, int n, const Point2f* Q, in
}
// Quit when both adv. indices have cycled, or one has cycled twice.
}
while ( ((aa < n) || (ba < m)) && (aa < 2*n) && (ba < 2*m) );
while ( ((aa < n) || (ba < m)) && (aa < 2*n) && (ba < 2*m) && ((int)(result - result0) <= result_size) );
// Deal with special cases: not implemented.
if( inflag == Unknown )
@@ -407,10 +410,16 @@ static int intersectConvexConvex_( const Point2f* P, int n, const Point2f* Q, in
// ...
}
int i, nr = (int)(result - result0);
int nr = (int)(result - result0);
if (nr > result_size)
{
*_area = -1.f;
return -1;
}
double area = 0;
Point2f prev = result0[nr-1];
for( i = 1; i < nr; i++ )
for(int i = 1; i < nr; i++ )
{
result0[i-1] = result0[i];
area += (double)prev.x*result0[i].y - (double)prev.y*result0[i].x;
@@ -445,9 +454,11 @@ float cv::intersectConvexConvex( InputArray _p1, InputArray _p2, OutputArray _p1
return 0.f;
}
AutoBuffer<Point2f> _result(n*2 + m*2 + 1);
Point2f *fp1 = _result.data(), *fp2 = fp1 + n;
AutoBuffer<Point2f> _result(n + m + n+m+1+3);
Point2f* fp1 = _result.data();
Point2f* fp2 = fp1 + n;
Point2f* result = fp2 + m;
int orientation = 0;
for( int k = 1; k <= 2; k++ )
@@ -476,7 +487,15 @@ float cv::intersectConvexConvex( InputArray _p1, InputArray _p2, OutputArray _p1
}
float area = 0.f;
int nr = intersectConvexConvex_(fp1, n, fp2, m, result, &area);
int nr = intersectConvexConvex_(fp1, n, fp2, m, result, n+m+1, &area);
if (nr < 0)
{
// The algorithm did not converge, e.g. some of inputs is not convex
_p12.release();
return -1.f;
}
if( nr == 0 )
{
if( !handleNested )
modules/imgproc/test/test_bilateral_filter.cpp
+1 -1
View File
@@ -243,7 +243,7 @@ namespace opencv_test { namespace {
rng.fill(_src, RNG::UNIFORM, 0, 256);
_sigma_color = _sigma_space = 1.;
_sigma_color = _sigma_space = rng.uniform(0., 10.);
return 1;
}
modules/imgproc/test/test_histograms.cpp
+69
View File
@@ -142,5 +142,74 @@ TEST(Imgproc_Hist_Calc, IPP_ranges_with_nonequal_exponent_21595)
ASSERT_EQ(histogram_u.at<float>(2), 4.f) << "1 not counts correctly, res: " << histogram_u.at<float>(2);
}
////////////////////////////////////////// equalizeHist() /////////////////////////////////////////
void equalizeHistReference(const Mat& src, Mat& dst)
{
std::vector<int> hist(256, 0);
for (int y = 0; y < src.rows; y++)
{
const uchar* srow = src.ptr(y);
for (int x = 0; x < src.cols; x++)
{
hist[srow[x]]++;
}
}
int first = 0;
while (!hist[first]) ++first;
int total = (int)src.total();
if (hist[first] == total)
{
dst.setTo(first);
return;
}
std::vector<uchar> lut(256);
lut[first] = 0;
float scale = (255.f)/(total - hist[first]);
int sum = 0;
for (int i = first + 1; i < 256; ++i)
{
sum += hist[i];
lut[i] = saturate_cast<uchar>(sum * scale);
}
cv::LUT(src, lut, dst);
}
typedef ::testing::TestWithParam<std::tuple<cv::Size, int>> Imgproc_Equalize_Hist;
TEST_P(Imgproc_Equalize_Hist, accuracy)
{
auto p = GetParam();
cv::Size size = std::get<0>(p);
int idx = std::get<1>(p);
RNG &rng = cvtest::TS::ptr()->get_rng();
rng.state += idx;
cv::Mat src(size, CV_8U);
cvtest::randUni(rng, src, Scalar::all(0), Scalar::all(255));
cv::Mat dst, gold;
equalizeHistReference(src, gold);
cv::equalizeHist(src, dst);
ASSERT_EQ(CV_8UC1, dst.type());
ASSERT_EQ(gold.size(), dst.size());
EXPECT_MAT_NEAR(dst, gold, 1);
EXPECT_MAT_N_DIFF(dst, gold, 0.05 * size.area()); // The 5% range could be accomodated to HAL
}
INSTANTIATE_TEST_CASE_P(Imgproc_Hist, Imgproc_Equalize_Hist, ::testing::Combine(
::testing::Values(cv::Size(123, 321), cv::Size(256, 256), cv::Size(1024, 768)),
::testing::Range(0, 10)));
}} // namespace
/* End Of File */
modules/imgproc/test/test_intersectconvexconvex.cpp
+36
View File
@@ -255,6 +255,42 @@ TEST(Imgproc_IntersectConvexConvex, intersection_4)
EXPECT_NEAR(area, 391500, std::numeric_limits<float>::epsilon());
}
// The inputs are not convex and cuased buffer overflow
// See https://github.com/opencv/opencv/issues/25259
TEST(Imgproc_IntersectConvexConvex, not_convex)
{
std::vector<cv::Point2f> convex1 = {
{ 46.077175f , 228.66121f }, { 5.428622f , 250.05899f }, {207.51741f , 109.645676f },
{175.94789f , 32.6566f }, {217.4915f , 252.66176f }, {187.09386f , 6.3988557f},
{ 52.20488f , 69.266205f }, { 38.188286f , 134.48068f }, {246.4742f , 31.41043f },
{178.97946f , 169.52287f }, {103.40764f , 153.30397f }, {160.67746f , 17.166115f },
{152.44255f , 135.35f }, {197.03804f , 193.04782f }, {248.28397f , 56.821487f },
{ 10.907227f , 82.55291f }, {109.67949f , 70.7405f }, { 58.96842f , 150.132f },
{150.7613f , 129.54753f }, {254.98463f , 228.21748f }, {139.02563f , 193.89336f },
{ 84.79946f , 162.25363f }, { 39.83567f , 44.626484f }, {107.034996f , 209.38887f },
{ 67.61073f , 17.119232f }, {208.8617f , 33.67367f }, {182.65207f , 8.291072f },
{ 72.89319f , 42.51845f }, {202.4902f , 123.97209f }, { 79.945076f , 140.99268f },
{225.8952f , 66.226326f }, { 34.08404f , 219.2208f }, {243.1221f , 60.95162f }
};
std::vector<cv::Point2f> convex2 = {
{144.33624f , 247.15732f }, { 5.656847f , 17.461054f }, {230.54338f , 2.0446582f},
{143.0578f , 215.27856f }, {250.44626f , 82.54287f }, { 0.3846766f, 11.101262f },
{ 70.81022f , 17.243904f }, { 77.18812f , 75.760666f }, {190.34933f , 234.30962f },
{230.10204f , 133.67998f }, { 58.903755f , 252.96451f }, {213.57228f , 155.7058f },
{190.80992f , 212.90802f }, {203.4356f , 36.55016f }, { 32.276424f , 2.5646307f},
{ 39.73823f , 87.23782f }, {112.46902f , 101.81753f }, { 58.154305f , 238.40395f },
{187.01064f , 96.24343f }, { 44.42692f , 10.573529f }, {118.76949f , 233.35114f },
{ 86.26109f , 120.93148f }, {217.94751f , 130.5933f }, {148.2687f , 68.56015f },
{187.44174f , 214.32857f }, {247.19875f , 180.8494f }, { 17.986013f , 61.451443f },
{254.74344f , 204.71747f }, {211.92726f , 132.0139f }, { 51.36624f , 116.63085f },
{ 83.80044f , 124.20074f }, {122.125854f , 25.182402f }, { 39.08164f , 180.08517f }
};
std::vector<cv::Point> intersection;
float area = cv::intersectConvexConvex(convex1, convex2, intersection, false);
EXPECT_TRUE(intersection.empty());
EXPECT_LE(area, 0.f);
}
} // namespace
} // opencv_test
modules/js/test/init_cv.js
+10 -5
View File
@@ -2,13 +2,18 @@
// It is subject to the license terms in the LICENSE file found in the top-level directory
// of this distribution and at http://opencv.org/license.html.
if (cv instanceof Promise) {
QUnit.test("init_cv", (assert) => {
QUnit.test("init_cv", (assert) => {
if (cv instanceof Promise) {
const done = assert.async();
assert.ok(true);
cv.then((ready_cv) => {
cv = ready_cv;
done();
});
});
}
} else if (cv.getBuildInformation === undefined) {
const done = assert.async();
cv['onRuntimeInitialized'] = () => {
done();
}
}
assert.ok(true);
});
modules/video/include/opencv2/video/tracking.hpp
+1
View File
@@ -879,6 +879,7 @@ public:
CV_PROP_RW int target;
CV_PROP_RW Scalar meanvalue;
CV_PROP_RW Scalar stdvalue;
CV_PROP_RW float tracking_score_threshold;
};
/** @brief Constructor
modules/video/src/tracking/tracker_vit.cpp
+34 -33
View File
@@ -24,8 +24,8 @@ TrackerVit::~TrackerVit()
TrackerVit::Params::Params()
{
net = "vitTracker.onnx";
meanvalue = Scalar{0.485, 0.456, 0.406};
stdvalue = Scalar{0.229, 0.224, 0.225};
meanvalue = Scalar{0.485, 0.456, 0.406}; // normalized mean (already divided by 255)
stdvalue = Scalar{0.229, 0.224, 0.225}; // normalized std (already divided by 255)
#ifdef HAVE_OPENCV_DNN
backend = dnn::DNN_BACKEND_DEFAULT;
target = dnn::DNN_TARGET_CPU;
@@ -33,6 +33,7 @@ TrackerVit::Params::Params()
backend = -1; // invalid value
target = -1; // invalid value
#endif
tracking_score_threshold = 0.20f; // safe threshold to filter out black frames
}
#ifdef HAVE_OPENCV_DNN
@@ -48,6 +49,9 @@ public:
net.setPreferableBackend(params.backend);
net.setPreferableTarget(params.target);
i2bp.mean = params.meanvalue * 255.0;
i2bp.scalefactor = (1.0 / params.stdvalue) * (1 / 255.0);
}
void init(InputArray image, const Rect& boundingBox) CV_OVERRIDE;
@@ -58,6 +62,7 @@ public:
float tracking_score;
TrackerVit::Params params;
dnn::Image2BlobParams i2bp;
protected:
@@ -69,10 +74,9 @@ protected:
Mat hanningWindow;
dnn::Net net;
Mat image;
};
static void crop_image(const Mat& src, Mat& dst, Rect box, int factor)
static int crop_image(const Mat& src, Mat& dst, Rect box, int factor)
{
int x = box.x, y = box.y, w = box.width, h = box.height;
int crop_sz = cvCeil(sqrt(w * h) * factor);
@@ -90,21 +94,16 @@ static void crop_image(const Mat& src, Mat& dst, Rect box, int factor)
Rect roi(x1 + x1_pad, y1 + y1_pad, x2 - x2_pad - x1 - x1_pad, y2 - y2_pad - y1 - y1_pad);
Mat im_crop = src(roi);
copyMakeBorder(im_crop, dst, y1_pad, y2_pad, x1_pad, x2_pad, BORDER_CONSTANT);
return crop_sz;
}
void TrackerVitImpl::preprocess(const Mat& src, Mat& dst, Size size)
{
Mat mean = Mat(size, CV_32FC3, params.meanvalue);
Mat std = Mat(size, CV_32FC3, params.stdvalue);
mean = dnn::blobFromImage(mean, 1.0, Size(), Scalar(), false);
std = dnn::blobFromImage(std, 1.0, Size(), Scalar(), false);
Mat img;
resize(src, img, size);
dst = dnn::blobFromImage(img, 1.0, Size(), Scalar(), false);
dst /= 255;
dst = (dst - mean) / std;
dst = dnn::blobFromImageWithParams(img, i2bp);
}
static Mat hann1d(int sz, bool centered = true) {
@@ -141,22 +140,21 @@ static Mat hann2d(Size size, bool centered = true) {
return hanningWindow;
}
static Rect returnfromcrop(float x, float y, float w, float h, Rect res_Last)
static void updateLastRect(float cx, float cy, float w, float h, int crop_size, Rect &rect_last)
{
int cropwindowwh = 4 * cvFloor(sqrt(res_Last.width * res_Last.height));
int x0 = res_Last.x + (res_Last.width - cropwindowwh) / 2;
int y0 = res_Last.y + (res_Last.height - cropwindowwh) / 2;
Rect finalres;
finalres.x = cvFloor(x * cropwindowwh + x0);
finalres.y = cvFloor(y * cropwindowwh + y0);
finalres.width = cvFloor(w * cropwindowwh);
finalres.height = cvFloor(h * cropwindowwh);
return finalres;
int x0 = rect_last.x + (rect_last.width - crop_size) / 2;
int y0 = rect_last.y + (rect_last.height - crop_size) / 2;
float x1 = cx - w / 2, y1 = cy - h / 2;
rect_last.x = cvFloor(x1 * crop_size + x0);
rect_last.y = cvFloor(y1 * crop_size + y0);
rect_last.width = cvFloor(w * crop_size);
rect_last.height = cvFloor(h * crop_size);
}
void TrackerVitImpl::init(InputArray image_, const Rect &boundingBox_)
{
image = image_.getMat().clone();
Mat image = image_.getMat();
Mat crop;
crop_image(image, crop, boundingBox_, 2);
Mat blob;
@@ -169,9 +167,9 @@ void TrackerVitImpl::init(InputArray image_, const Rect &boundingBox_)
bool TrackerVitImpl::update(InputArray image_, Rect &boundingBoxRes)
{
image = image_.getMat().clone();
Mat image = image_.getMat();
Mat crop;
crop_image(image, crop, rect_last, 4);
int crop_size = crop_image(image, crop, rect_last, 4); // crop: [crop_size, crop_size]
Mat blob;
preprocess(crop, blob, searchSize);
net.setInput(blob, "search");
@@ -191,15 +189,18 @@ bool TrackerVitImpl::update(InputArray image_, Rect &boundingBoxRes)
minMaxLoc(conf_map, nullptr, &maxVal, nullptr, &maxLoc);
tracking_score = static_cast<float>(maxVal);
float cx = (maxLoc.x + offset_map.at<float>(0, maxLoc.y, maxLoc.x)) / 16;
float cy = (maxLoc.y + offset_map.at<float>(1, maxLoc.y, maxLoc.x)) / 16;
float w = size_map.at<float>(0, maxLoc.y, maxLoc.x);
float h = size_map.at<float>(1, maxLoc.y, maxLoc.x);
if (tracking_score >= params.tracking_score_threshold) {
float cx = (maxLoc.x + offset_map.at<float>(0, maxLoc.y, maxLoc.x)) / 16;
float cy = (maxLoc.y + offset_map.at<float>(1, maxLoc.y, maxLoc.x)) / 16;
float w = size_map.at<float>(0, maxLoc.y, maxLoc.x);
float h = size_map.at<float>(1, maxLoc.y, maxLoc.x);
Rect finalres = returnfromcrop(cx - w / 2, cy - h / 2, w, h, rect_last);
rect_last = finalres;
boundingBoxRes = finalres;
return true;
updateLastRect(cx, cy, w, h, crop_size, rect_last);
boundingBoxRes = rect_last;
return true;
} else {
return false;
}
}
float TrackerVitImpl::getTrackingScore()
modules/video/test/test_trackers.cpp
+1 -3
View File
@@ -143,9 +143,7 @@ TEST(vittrack, accuracy_vittrack)
cv::TrackerVit::Params params;
params.net = model;
cv::Ptr<Tracker> tracker = TrackerVit::create(params);
// NOTE: Test threshold was reduced from 0.67 (libjpeg-turbo) to 0.66 (libjpeg 9f),
// becase libjpeg and libjpeg-turbo produce slightly different images
checkTrackingAccuracy(tracker, 0.66);
checkTrackingAccuracy(tracker, 0.64);
}
}} // namespace opencv_test::
modules/videoio/CMakeLists.txt
+4
View File
@@ -49,6 +49,9 @@ endif()
# Removing WinRT API headers by default
list(REMOVE_ITEM videoio_ext_hdrs "${CMAKE_CURRENT_LIST_DIR}/include/opencv2/${name}/cap_winrt.hpp")
# Remove iOS API header by default
list(REMOVE_ITEM videoio_ext_hdrs "${CMAKE_CURRENT_LIST_DIR}/include/opencv2/${name}/cap_ios.h")
if(DEFINED WINRT AND NOT DEFINED ENABLE_WINRT_MODE_NATIVE)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /ZW")
endif()
@@ -221,6 +224,7 @@ if(TARGET ocv.3rdparty.cap_ios)
${CMAKE_CURRENT_LIST_DIR}/src/cap_ios_photo_camera.mm
${CMAKE_CURRENT_LIST_DIR}/src/cap_ios_video_camera.mm)
list(APPEND tgts ocv.3rdparty.cap_ios)
list(APPEND videoio_ext_hdrs "${CMAKE_CURRENT_LIST_DIR}/include/opencv2/${name}/cap_ios.h")
endif()
if(TARGET ocv.3rdparty.android_mediandk)
modules/videoio/src/cap_gstreamer.cpp
+1 -1
View File
@@ -1618,7 +1618,7 @@ bool GStreamerCapture::open(const String &filename_, const cv::VideoCaptureParam
{
//do not emit signals: all calls will be synchronous and blocking
gst_app_sink_set_emit_signals (GST_APP_SINK(sink.get()), FALSE);
caps.attach(gst_caps_from_string("video/x-raw, format=(string){BGR, GRAY8}; video/x-bayer,format=(string){rggb,bggr,grbg,gbrg}; image/jpeg"));
caps.attach(gst_caps_from_string("video/x-raw, format=(string){BGR}; video/x-raw, format=(string){BGRx, BGRA}; video/x-bayer,format=(string){rggb,bggr,grbg,gbrg}; video/x-raw, format=(string){GRAY8}; image/jpeg"));
}
if (audioStream >= 0)
{
platforms/android/build_sdk.py
+7 -3
View File
@@ -267,10 +267,14 @@ class Builder:
cmake_vars['WITH_ANDROID_MEDIANDK'] = "OFF"
if self.hwasan and "arm64" in abi.name:
cmake_vars['OPENCV_ENABLE_MEMORY_SANITIZER'] = "ON"
hwasan_flags = "-fno-omit-frame-pointer -fsanitize=hwaddress"
cmake_vars['CMAKE_CXX_FLAGS_DEBUG'] = hwasan_flags
cmake_vars['CMAKE_C_FLAGS_DEBUG'] = hwasan_flags
cmake_vars['CMAKE_LINKER_FLAGS_DEBUG'] = hwasan_flags
for s in ['OPENCV_EXTRA_C_FLAGS', 'OPENCV_EXTRA_CXX_FLAGS', 'OPENCV_EXTRA_EXE_LINKER_FLAGS',
'OPENCV_EXTRA_SHARED_LINKER_FLAGS', 'OPENCV_EXTRA_MODULE_LINKER_FLAGS']:
if s in cmake_vars.keys():
cmake_vars[s] = cmake_vars[s] + ' ' + hwasan_flags
else:
cmake_vars[s] = hwasan_flags
cmake_vars.update(abi.cmake_vars)
samples/cpp/tutorial_code/videoio/orbbec_astra/orbbec_astra.cpp → samples/cpp/tutorial_code/videoio/openni_orbbec_astra/openni_orbbec_astra.cpp
View File
samples/dnn/download_models.py
-1
View File
@@ -3,7 +3,6 @@ Helper module to download extra data from Internet
'''
from __future__ import print_function
import os
import cv2
import sys
import yaml
import argparse
samples/dnn/models.yml
+14
View File
@@ -62,6 +62,20 @@ yolov8m:
background_label_id: 0
sample: "yolo_detector"
yolov8l:
load_info:
url: "https://github.com/CVHub520/X-AnyLabeling/releases/download/v0.1.0/yolov8l.onnx"
sha1: "462df53ca3a85d110bf6be7fc2e2bb1277124395"
model: "yolov8l.onnx"
mean: 0.0
scale: 0.00392
width: 640
height: 640
rgb: true
classes: "object_detection_classes_yolo.txt"
background_label_id: 0
sample: "yolo_detector"
# YOLO4 object detection family from Darknet (https://github.com/AlexeyAB/darknet)
# YOLO object detection family from Darknet (https://pjreddie.com/darknet/yolo/)
# Might be used for all YOLOv2, TinyYolov2, YOLOv3, YOLOv4 and TinyYolov4
samples/dnn/vit_tracker.cpp
+21 -14
View File
@@ -16,6 +16,7 @@ const char *keys =
"{ help h | | Print help message }"
"{ input i | | Full path to input video folder, the specific camera index. (empty for camera 0) }"
"{ net | vitTracker.onnx | Path to onnx model of vitTracker.onnx}"
"{ tracking_score_threshold t | 0.3 | Tracking score threshold. If a bbox of score >= 0.3, it is considered as found }"
"{ backend | 0 | Choose one of computation backends: "
"0: automatically (by default), "
"1: Halide language (http://halide-lang.org/), "
@@ -49,6 +50,7 @@ int run(int argc, char** argv)
std::string net = parser.get<String>("net");
int backend = parser.get<int>("backend");
int target = parser.get<int>("target");
float tracking_score_threshold = parser.get<float>("tracking_score_threshold");
Ptr<TrackerVit> tracker;
try
@@ -57,6 +59,7 @@ int run(int argc, char** argv)
params.net = samples::findFile(net);
params.backend = backend;
params.target = target;
params.tracking_score_threshold = tracking_score_threshold;
tracker = TrackerVit::create(params);
}
catch (const cv::Exception& ee)
@@ -108,6 +111,11 @@ int run(int argc, char** argv)
Rect selectRect = selectROI(winName, image_select);
std::cout << "ROI=" << selectRect << std::endl;
if (selectRect.empty())
{
std::cerr << "Invalid ROI!" << std::endl;
return 2;
}
tracker->init(image, selectRect);
@@ -130,30 +138,29 @@ int run(int argc, char** argv)
float score = tracker->getTrackingScore();
std::cout << "frame " << count <<
": predicted score=" << score <<
" rect=" << rect <<
" time=" << tickMeter.getTimeMilli() << "ms" <<
std::endl;
std::cout << "frame " << count;
if (ok) {
std::cout << ": predicted score=" << score <<
"\trect=" << rect <<
"\ttime=" << tickMeter.getTimeMilli() << "ms" << std::endl;
Mat render_image = image.clone();
if (ok)
{
rectangle(render_image, rect, Scalar(0, 255, 0), 2);
rectangle(image, rect, Scalar(0, 255, 0), 2);
std::string timeLabel = format("Inference time: %.2f ms", tickMeter.getTimeMilli());
std::string scoreLabel = format("Score: %f", score);
putText(render_image, timeLabel, Point(0, 15), FONT_HERSHEY_SIMPLEX, 0.5, Scalar(0, 255, 0));
putText(render_image, scoreLabel, Point(0, 35), FONT_HERSHEY_SIMPLEX, 0.5, Scalar(0, 255, 0));
putText(image, timeLabel, Point(0, 15), FONT_HERSHEY_SIMPLEX, 0.5, Scalar(0, 255, 0));
putText(image, scoreLabel, Point(0, 35), FONT_HERSHEY_SIMPLEX, 0.5, Scalar(0, 255, 0));
} else {
std::cout << ": target lost" << std::endl;
putText(image, "Target lost", Point(0, 15), FONT_HERSHEY_SIMPLEX, 0.5, Scalar(0, 0, 255));
}
imshow(winName, render_image);
imshow(winName, image);
tickMeter.reset();
int c = waitKey(1);
if (c == 27 /*ESC*/)
if (c == 27 /*ESC*/ || c == 'q' || c == 'Q')
break;
}