platform/upstream/opencv.git
7 years agosobel and scharr ocl kernel optimization
Li Peng [Thu, 27 Oct 2016 04:41:40 +0000 (12:41 +0800)]
sobel and scharr ocl kernel optimization

It improves 108%~230% performance in the perf test
with image format "CV_8UC1" and kernel size 3.

Signed-off-by: Li Peng <peng.li@intel.com>
7 years agoMerge pull request #7628 from sergiud:viz-vtkUnsignedCharArray-delete
Alexander Alekhin [Thu, 10 Nov 2016 17:04:04 +0000 (17:04 +0000)]
Merge pull request #7628 from sergiud:viz-vtkUnsignedCharArray-delete

7 years agoMerge pull request #7629 from alalek:issue_7626
Alexander Alekhin [Thu, 10 Nov 2016 16:35:08 +0000 (16:35 +0000)]
Merge pull request #7629 from alalek:issue_7626

7 years agoMerge pull request #7161 from terfendail:shortline_fix
Vadim Pisarevsky [Thu, 10 Nov 2016 16:02:19 +0000 (16:02 +0000)]
Merge pull request #7161 from terfendail:shortline_fix

7 years agoMerge pull request #7556 from terfendail:ovxhal_101compl
Vadim Pisarevsky [Thu, 10 Nov 2016 15:56:01 +0000 (15:56 +0000)]
Merge pull request #7556 from terfendail:ovxhal_101compl

7 years agoMerge pull request #7550 from apavlenko:ivx
Andrey Pavlenko [Thu, 10 Nov 2016 11:43:08 +0000 (15:43 +0400)]
Merge pull request #7550 from apavlenko:ivx

C++ wrappers over OpenVX 1.x API (#7550)

* initial commit

* fixing missing swap(refcount) bug

* removing dead code

* fixinf identation

* splitting c++98 and external refcount definitions

* fixing formatToMatType

* using vector instead of initializer_list for consistency with c++98 API

* IVX_HIDE_INFO_WARNINGS

* just in case a compiler doesn't do this

* removing copy functionality from ImagePatch, Image can be used for that

* stricter type checks

7 years agoOpenVX 1.0.1 compliance update
Vitaly Tuzov [Wed, 26 Oct 2016 12:33:08 +0000 (15:33 +0300)]
OpenVX 1.0.1 compliance update

7 years agoMerge pull request #7632 from paroj:homgpt_int
Alexander Alekhin [Wed, 9 Nov 2016 14:55:10 +0000 (14:55 +0000)]
Merge pull request #7632 from paroj:homgpt_int

7 years agoMerge pull request #7602 from mshabunin:fix-opencl-warnings
Alexander Alekhin [Wed, 9 Nov 2016 12:35:00 +0000 (12:35 +0000)]
Merge pull request #7602 from mshabunin:fix-opencl-warnings

7 years agoconvertPointsToHomogeneous: fix output type for depth == CV_32S
Pavel Rojtberg [Wed, 9 Nov 2016 10:35:12 +0000 (11:35 +0100)]
convertPointsToHomogeneous: fix output type for depth == CV_32S

7 years agoMerge pull request #7624 from pengli:gaussian_blur
Alexander Alekhin [Wed, 9 Nov 2016 09:25:25 +0000 (09:25 +0000)]
Merge pull request #7624 from pengli:gaussian_blur

7 years agoMerge pull request #7618 from lupustr3:pvlasov/instrumentation_update
Alexander Alekhin [Tue, 8 Nov 2016 15:37:03 +0000 (15:37 +0000)]
Merge pull request #7618 from lupustr3:pvlasov/instrumentation_update

7 years agoimgproc: fix trailingZeros for MSVS 2010
Alexander Alekhin [Tue, 8 Nov 2016 12:44:52 +0000 (15:44 +0300)]
imgproc: fix trailingZeros for MSVS 2010

7 years agoviz: free new[] allocated arrays using delete[] instead of free
Sergiu Deitsch [Tue, 8 Nov 2016 11:54:58 +0000 (12:54 +0100)]
viz: free new[] allocated arrays using delete[] instead of free

Clang's AddressSanitizer throws an alloc-dealloc-mismatch (operator
new[] vs free) error here.

7 years ago--perf_instrument parameter now has int type and 0, 1, 2 modes (1 - simple trees...
Pavel Vlasov [Mon, 7 Nov 2016 09:15:51 +0000 (12:15 +0300)]
--perf_instrument parameter now has int type and 0, 1, 2 modes (1 - simple trees, 2 - expanded trees for functions with same name but different calling address);

Maximum depth limit var was added to the instrumentation structure;

Trace names output console output fix: improper tree formatting could happen;
Output in case of error was added;

Custom regions improvements;

Improved timing and weight calculation for parallel regions; New TC (threads counter) value to indicate how many different threads accessed particular node;

parallel_for, warnings fixes and ReturnAddress code from Alexander Alekhin;

7 years agogaussian blur ocl kernel optimization
Li Peng [Fri, 21 Oct 2016 05:17:45 +0000 (13:17 +0800)]
gaussian blur ocl kernel optimization

This ocl kernel is for 3x3 kernel size and CV_8UC1 format
It is 115% ~ 300% faster than current ocl path in perf test

python ./modules/ts/misc/run.py -t imgproc --gtest_filter=OCL_GaussianBlurFixture*

Signed-off-by: Li Peng <peng.li@intel.com>
7 years agoMerge pull request #7585 from pengli:morph_filter
Alexander Alekhin [Mon, 7 Nov 2016 17:11:32 +0000 (17:11 +0000)]
Merge pull request #7585 from pengli:morph_filter

7 years agoFixed several OpenCL compiler warnings
mshabunin [Wed, 2 Nov 2016 11:59:37 +0000 (14:59 +0300)]
Fixed several OpenCL compiler warnings

7 years agomorph ocl kernel for erode and dilate filter
Li Peng [Wed, 19 Oct 2016 06:53:05 +0000 (14:53 +0800)]
morph ocl kernel for erode and dilate filter

This kernel is for CV_8UC1 format and 3x3 kernel size,
It is about 33% ~ 55% faster than current ocl kernel with below perf test

python ./modules/ts/misc/run.py -t imgproc --gtest_filter=OCL_ErodeFixture*
python ./modules/ts/misc/run.py -t imgproc --gtest_filter=OCL_DilateFixture*

Also add accuracy test cases for this kernel, the test command is

./bin/opencv_test_imgproc --gtest_filter=OCL_Filter/MorphFilter3x3*

Signed-off-by: Li Peng <peng.li@intel.com>
7 years agoMerge pull request #7604 from rokm:highgui-qt
Alexander Alekhin [Thu, 3 Nov 2016 19:24:52 +0000 (19:24 +0000)]
Merge pull request #7604 from rokm:highgui-qt

7 years agoMerge pull request #7423 from grahamfyffe:leading_edges
Vadim Pisarevsky [Thu, 3 Nov 2016 11:20:42 +0000 (11:20 +0000)]
Merge pull request #7423 from grahamfyffe:leading_edges

7 years agoMerge pull request #7605 from tomoaki0705:fixBuildFp16Error
Alexander Alekhin [Thu, 3 Nov 2016 07:41:42 +0000 (07:41 +0000)]
Merge pull request #7605 from tomoaki0705:fixBuildFp16Error

7 years agoMerge pull request #7595 from sovrasov:fs_utf8_bom
Vadim Pisarevsky [Wed, 2 Nov 2016 15:27:48 +0000 (15:27 +0000)]
Merge pull request #7595 from sovrasov:fs_utf8_bom

7 years agofix build error
Tomoaki Teshima [Wed, 2 Nov 2016 14:18:58 +0000 (23:18 +0900)]
fix build error

7 years agoMerge pull request #7601 from sovrasov:remap_big_img_assert
Vadim Pisarevsky [Wed, 2 Nov 2016 14:17:04 +0000 (14:17 +0000)]
Merge pull request #7601 from sovrasov:remap_big_img_assert

7 years agoMerge pull request #7596 from sovrasov:doc_core_clarify
Vadim Pisarevsky [Wed, 2 Nov 2016 13:32:14 +0000 (13:32 +0000)]
Merge pull request #7596 from sovrasov:doc_core_clarify

7 years agoMerge pull request #7467 from tomoaki0705:featureCheckSimdUniversal
Vadim Pisarevsky [Wed, 2 Nov 2016 12:30:55 +0000 (12:30 +0000)]
Merge pull request #7467 from tomoaki0705:featureCheckSimdUniversal

7 years agoMerge pull request #7552 from es0m:master
Vadim Pisarevsky [Wed, 2 Nov 2016 12:12:09 +0000 (12:12 +0000)]
Merge pull request #7552 from es0m:master

7 years agohighgui: window_QT: added missing virtual method specifier
Rok Mandeljc [Wed, 2 Nov 2016 12:11:51 +0000 (13:11 +0100)]
highgui: window_QT: added missing virtual method specifier

Fixes #7603, which was caused by OCVViewPort::icvmouseProcessing
not being declared as virtual, and hence was not overriden by
DefaultViewPort::icvmouseProcessing (which does the inverse
coordinate mapping).

7 years agoAdd new error messages
Vladislav Sovrasov [Wed, 2 Nov 2016 11:53:04 +0000 (14:53 +0300)]
Add new error messages

7 years agoSkip UTF-8 BOM in FileStorage
Vladislav Sovrasov [Tue, 1 Nov 2016 09:24:30 +0000 (12:24 +0300)]
Skip UTF-8 BOM in FileStorage

7 years agoAdd assertion to prevent processing of large images in remap
Vladislav Sovrasov [Wed, 2 Nov 2016 11:17:56 +0000 (14:17 +0300)]
Add assertion to prevent processing of large images in remap

7 years agoClarify documentation for cv::Algorithm::read
Vladislav Sovrasov [Tue, 1 Nov 2016 13:06:08 +0000 (16:06 +0300)]
Clarify documentation for cv::Algorithm::read

7 years agoMerge pull request #7526 from alalek:fix_arm_builds
Vadim Pisarevsky [Wed, 2 Nov 2016 10:49:43 +0000 (10:49 +0000)]
Merge pull request #7526 from alalek:fix_arm_builds

7 years agoMerge pull request #7582 from K-Shinotsuka:issue32
Vadim Pisarevsky [Wed, 2 Nov 2016 10:48:29 +0000 (10:48 +0000)]
Merge pull request #7582 from K-Shinotsuka:issue32

7 years agoMerge pull request #7583 from K-Shinotsuka:issue33
Vadim Pisarevsky [Wed, 2 Nov 2016 10:47:33 +0000 (10:47 +0000)]
Merge pull request #7583 from K-Shinotsuka:issue33

7 years agoMerge pull request #7584 from tomoaki0705:fixFp16BuildOldCompiler
Vadim Pisarevsky [Wed, 2 Nov 2016 10:43:34 +0000 (10:43 +0000)]
Merge pull request #7584 from tomoaki0705:fixFp16BuildOldCompiler

7 years agoadd SSE code for Lab2RGB_f.
k-shinotsuka [Sun, 30 Oct 2016 08:31:59 +0000 (17:31 +0900)]
add SSE code for Lab2RGB_f.

7 years agoMerge pull request #7571 from panxiaochun:CvVideoCamera_stop
Alexander Alekhin [Mon, 31 Oct 2016 17:35:41 +0000 (17:35 +0000)]
Merge pull request #7571 from panxiaochun:CvVideoCamera_stop

7 years agoMerge pull request #7497 from abratchik:java.wrapper.fix.3.1
Alexander Alekhin [Mon, 31 Oct 2016 17:24:18 +0000 (17:24 +0000)]
Merge pull request #7497 from abratchik:java.wrapper.fix.3.1

7 years agofix build error on old compiler
Tomoaki Teshima [Mon, 31 Oct 2016 14:20:44 +0000 (23:20 +0900)]
fix build error on old compiler
  * vget_lane_** accepts constant value, not variable

7 years agofixed _MSC_VER check for at least vs2013 in dbt_face_detection sample
Eric Sommerlade [Mon, 31 Oct 2016 13:48:41 +0000 (13:48 +0000)]
fixed _MSC_VER check for at least vs2013 in dbt_face_detection sample

7 years agofixed _MSC_VER check for vs2013 in dbt_face_detection sample
Eric Sommerlade [Mon, 31 Oct 2016 13:46:22 +0000 (13:46 +0000)]
fixed _MSC_VER check for vs2013 in dbt_face_detection sample

7 years agoMerge pull request #7589 from alalek:fix_perf_stitching
Alexander Alekhin [Mon, 31 Oct 2016 12:10:13 +0000 (12:10 +0000)]
Merge pull request #7589 from alalek:fix_perf_stitching

7 years agoMerge branch 'master' of https://github.com/es0m/opencv
Eric Sommerlade [Mon, 31 Oct 2016 11:10:01 +0000 (11:10 +0000)]
Merge branch 'master' of https://github.com/es0m/opencv

7 years agofixed detection_based_tracker on visual studio 2013 and later
Eric Sommerlade [Mon, 31 Oct 2016 11:07:09 +0000 (11:07 +0000)]
fixed detection_based_tracker on visual studio 2013 and later
modules/objectdetect/src/detection_based_tracker.cpp: made unique_lock<mutex> local to each function
samples/cpp/dbt_face_detection.cpp: fixed warnings on loop in Visual Studio

7 years agotest: don't check magic values in stitching perf test
Alexander Alekhin [Mon, 31 Oct 2016 10:36:19 +0000 (13:36 +0300)]
test: don't check magic values in stitching perf test

7 years agoreversed mysterious mode change
E Sommerlade [Mon, 31 Oct 2016 09:47:44 +0000 (09:47 +0000)]
reversed mysterious mode change

7 years agofix for #7510, #7511 and #7512
abratchik [Tue, 18 Oct 2016 18:22:18 +0000 (22:22 +0400)]
fix for #7510, #7511 and #7512

7 years agofix java wrapper autotests
abratchik [Mon, 17 Oct 2016 17:53:59 +0000 (21:53 +0400)]
fix java wrapper autotests

7 years agoMerge pull request #7516 from sovrasov:find_contours_fix
Alexander Alekhin [Sun, 30 Oct 2016 08:33:11 +0000 (08:33 +0000)]
Merge pull request #7516 from sovrasov:find_contours_fix

7 years agoMerge pull request #7574 from Tytan:expcomp_before_seams
Alexander Alekhin [Sun, 30 Oct 2016 08:28:35 +0000 (08:28 +0000)]
Merge pull request #7574 from Tytan:expcomp_before_seams

7 years agoadd SSE code for Luv2RGB_f.
k-shinotsuka [Sun, 30 Oct 2016 03:02:39 +0000 (12:02 +0900)]
add SSE code for Luv2RGB_f.

7 years agoMerge pull request #7580 from seanlis:wrong_diag_comment
Vadim Pisarevsky [Sat, 29 Oct 2016 21:30:26 +0000 (21:30 +0000)]
Merge pull request #7580 from seanlis:wrong_diag_comment

7 years agoMerge pull request #7540 from K-Shinotsuka:issue31
Vadim Pisarevsky [Sat, 29 Oct 2016 21:28:25 +0000 (21:28 +0000)]
Merge pull request #7540 from K-Shinotsuka:issue31

7 years agoMerge pull request #7503 from pengli:box_filter_v2
Vadim Pisarevsky [Sat, 29 Oct 2016 21:20:06 +0000 (21:20 +0000)]
Merge pull request #7503 from pengli:box_filter_v2

7 years agoMerge pull request #7549 from StevenPuttemans:add_sample_content
Vadim Pisarevsky [Sat, 29 Oct 2016 21:13:37 +0000 (21:13 +0000)]
Merge pull request #7549 from StevenPuttemans:add_sample_content

7 years agoCorrect the comment of Mat::diag(const Mat& d).
Sean Li [Sat, 29 Oct 2016 08:01:23 +0000 (16:01 +0800)]
Correct the comment of Mat::diag(const Mat& d).

7 years agoadd SSE code for RGB2Luv_f.
k-shinotsuka [Sat, 22 Oct 2016 10:56:55 +0000 (19:56 +0900)]
add SSE code for RGB2Luv_f.

7 years agoadd universal hardware support check function
Tomoaki Teshima [Sat, 29 Oct 2016 04:24:31 +0000 (13:24 +0900)]
add universal hardware support check function
  * use hasSIMD128 rather than calling checkHardwareSupport
  * add SIMD check in spartialgradient.cpp
  * add SIMD check in stereosgbm.cpp
  * add SIMD check in canny.cpp

7 years agoMerge pull request #7572 from tomoaki0705:featureUniversalStereoSgbm
Vadim Pisarevsky [Fri, 28 Oct 2016 15:34:11 +0000 (15:34 +0000)]
Merge pull request #7572 from tomoaki0705:featureUniversalStereoSgbm

7 years agoCompensate exposure before finding seams
Quentin [Wed, 12 Oct 2016 11:53:06 +0000 (13:53 +0200)]
Compensate exposure before finding seams

7 years agoadd universal intrinsic in StereoSGBM
Tomoaki Teshima [Fri, 28 Oct 2016 12:47:13 +0000 (21:47 +0900)]
add universal intrinsic in StereoSGBM
  * add 8 elements version of reduce operation
  * add tests for new universal intrinsic

7 years agoMerge pull request #7561 from paroj:winvisible
Alexander Alekhin [Fri, 28 Oct 2016 11:31:04 +0000 (11:31 +0000)]
Merge pull request #7561 from paroj:winvisible

7 years agohighgui: add property WND_PROP_VISIBLE
Pavel Rojtberg [Thu, 27 Oct 2016 14:22:42 +0000 (16:22 +0200)]
highgui: add property WND_PROP_VISIBLE

checks whether the window exists and is visible. On QT closing a window
merley hides it, so the common hack for checking whether a window exists

exists = cv2.getWindowProperty(.., 0) >= 0

does not work.

7 years agoDisable redundant border initialization in findContours
Vladislav Sovrasov [Tue, 25 Oct 2016 09:20:36 +0000 (12:20 +0300)]
Disable redundant border initialization in findContours

7 years agofix the problem:repeatedly stop cause app crash
panxiaochun [Fri, 28 Oct 2016 07:50:37 +0000 (15:50 +0800)]
fix the problem:repeatedly stop cause app crash

7 years agoremove self. in release method
panxiaochun [Fri, 28 Oct 2016 06:48:49 +0000 (14:48 +0800)]
remove self. in release method

7 years agofix the problem: leak memory when repeatedly start and stop
panxiaochun [Fri, 28 Oct 2016 03:18:33 +0000 (11:18 +0800)]
fix the problem: leak memory when repeatedly start and stop

7 years agoMerge pull request #7554 from lupustr3:pvlasov/morph_isolated
Alexander Alekhin [Thu, 27 Oct 2016 16:51:03 +0000 (16:51 +0000)]
Merge pull request #7554 from lupustr3:pvlasov/morph_isolated

7 years agoMerge pull request #7558 from alalek:cv_xadd
Alexander Alekhin [Thu, 27 Oct 2016 15:59:17 +0000 (15:59 +0000)]
Merge pull request #7558 from alalek:cv_xadd

7 years agocore: update CV_XADD
Alexander Alekhin [Thu, 27 Oct 2016 10:22:13 +0000 (13:22 +0300)]
core: update CV_XADD

Allow to use custom implementations of CV_XADD (passed via -D of CXXFLAGS)

7 years agoMerge pull request #7553 from erichkeane:interlocked_add_fix
Alexander Alekhin [Thu, 27 Oct 2016 07:03:18 +0000 (07:03 +0000)]
Merge pull request #7553 from erichkeane:interlocked_add_fix

7 years agoBORDER_ISOLATED support for Laplacian, boxFilter and morphology;
Pavel Vlasov [Wed, 26 Oct 2016 11:07:41 +0000 (14:07 +0300)]
BORDER_ISOLATED support for Laplacian, boxFilter and morphology;

7 years agoremoved spaces at line endings, fixed warning
E Sommerlade [Wed, 26 Oct 2016 18:59:10 +0000 (19:59 +0100)]
removed spaces at line endings, fixed warning

7 years agoRemove Intel Specific CV_XADD Definition
Erich Keane [Wed, 26 Oct 2016 16:37:25 +0000 (09:37 -0700)]
Remove Intel Specific CV_XADD Definition

Intel supports __atomic_fetch_add, so it isn't necessary to use
a specific version otherwise.

7 years agorestored Parameters() constructor to maintain ABI compatibility. Added CV_EXPORTS...
E Sommerlade [Wed, 26 Oct 2016 14:59:44 +0000 (15:59 +0100)]
restored Parameters() constructor to maintain ABI compatibility. Added CV_EXPORTS to nested class

7 years agofixed use of std::lock outside of ifdefs
E Sommerlade [Wed, 26 Oct 2016 14:37:06 +0000 (15:37 +0100)]
fixed use of std::lock outside of ifdefs

7 years agoMerge pull request #7551 from fegorsch:fix_yaml_header
Alexander Alekhin [Wed, 26 Oct 2016 13:30:13 +0000 (13:30 +0000)]
Merge pull request #7551 from fegorsch:fix_yaml_header

7 years agoupdating python tutorials + providing necessary data
StevenPuttemans [Tue, 25 Oct 2016 09:41:03 +0000 (11:41 +0200)]
updating python tutorials + providing necessary data

7 years agoocl kernel performance optimization for box filter
Li Peng [Fri, 14 Oct 2016 07:13:10 +0000 (15:13 +0800)]
ocl kernel performance optimization for box filter

The optimization is for CV_8UC1 format and 3x3 box filter,
it is 15%~87% faster than current ocl kernel with below perf test

./modules/ts/misc/run.py -t imgproc --gtest_filter=OCL_BlurFixture*

Also add test cases for this ocl kernel.

Signed-off-by: Li Peng <peng.li@intel.com>
7 years agoCorrect cast for _InterlockedExchangeAdd on ICC
Erich Keane [Tue, 25 Oct 2016 21:18:55 +0000 (14:18 -0700)]
Correct cast for _InterlockedExchangeAdd on ICC

A bug in ICC improperly identified the first parameter as "void*"
rather than the proper "volatile long*".  This is scheduled to be
fixed in ICC in a future release.

This patch casts only to a "long*" to preserve backwards compatibility
with the ICC 16 and ICC 17 releases.

7 years agofixed detection_based_tracker on visual studio 2013 and later
E Sommerlade [Tue, 25 Oct 2016 19:18:41 +0000 (20:18 +0100)]
fixed detection_based_tracker on visual studio 2013 and later

7 years agoMerge pull request #2 from opencv/master
Eric Sommerlade [Tue, 25 Oct 2016 17:47:12 +0000 (18:47 +0100)]
Merge pull request #2 from opencv/master

update to master

7 years agoUse colon in "%YAML:1.0" directive when persisting to YAML-files
fegorsch [Tue, 25 Oct 2016 16:18:35 +0000 (18:18 +0200)]
Use colon in "%YAML:1.0" directive when persisting to YAML-files

In YAML 1.0 the colon is mandatory. See http://yaml.org/spec/1.0/#id2558635.

This also allows prior releases to read YAML files created with the current version.

7 years agoMerge pull request #7545 from sovrasov:flood_fill_fix_mask_value
Alexander Alekhin [Tue, 25 Oct 2016 14:24:19 +0000 (14:24 +0000)]
Merge pull request #7545 from sovrasov:flood_fill_fix_mask_value

7 years agoEnable chessboard detector test on Ubuntu
Vladislav Sovrasov [Tue, 25 Oct 2016 08:53:53 +0000 (11:53 +0300)]
Enable chessboard detector test on Ubuntu

7 years agoAdd a regression test, fix documentation
Vladislav Sovrasov [Wed, 19 Oct 2016 10:51:37 +0000 (13:51 +0300)]
Add a regression test, fix documentation

7 years agofindContours: integrate workaroung to allow contours detection on image border
Vladislav Sovrasov [Tue, 18 Oct 2016 11:16:14 +0000 (14:16 +0300)]
findContours: integrate workaroung to allow contours detection on image border

7 years agoFix wrong default mask value in floodFill
Vladislav Sovrasov [Mon, 24 Oct 2016 14:22:44 +0000 (17:22 +0300)]
Fix wrong default mask value in floodFill

7 years agoMerge pull request #7541 from tomoaki0705:fixArmDivCaroteneSaturate
Alexander Alekhin [Mon, 24 Oct 2016 15:35:05 +0000 (15:35 +0000)]
Merge pull request #7541 from tomoaki0705:fixArmDivCaroteneSaturate

7 years agobrush up divSaturate of carotene
Tomoaki Teshima [Sat, 22 Oct 2016 23:51:25 +0000 (08:51 +0900)]
brush up divSaturate of carotene
  * use rounding to nearest same as divSaturateQ

7 years agoMerge pull request #7517 from paroj:pp_dpdf
Alexander Alekhin [Sat, 22 Oct 2016 16:16:24 +0000 (16:16 +0000)]
Merge pull request #7517 from paroj:pp_dpdf

7 years agoMerge pull request #7531 from dtmoodie:cmake_updates
Alexander Alekhin [Sat, 22 Oct 2016 16:13:11 +0000 (16:13 +0000)]
Merge pull request #7531 from dtmoodie:cmake_updates

7 years agoMerge pull request #6933 from hrnr:gsoc_all
Jiri Horner [Sat, 22 Oct 2016 16:10:42 +0000 (18:10 +0200)]
Merge pull request #6933 from hrnr:gsoc_all

[GSOC] New camera model for stitching pipeline

* implement estimateAffine2D

estimates affine transformation using robust RANSAC method.

* uses RANSAC framework in calib3d
* includes accuracy test
* uses SVD decomposition for solving 3 point equation

* implement estimateAffinePartial2D

estimates limited affine transformation

* includes accuracy test

* stitching: add affine matcher

initial version of matcher that estimates affine transformation

* stitching: added affine transform estimator

initial version of estimator that simply chain transformations in homogeneous coordinates

* calib3d: rename estimateAffine3D test

test Calib3d_EstimateAffineTransform rename to Calib3d_EstimateAffine3D. This is more descriptive and prevents confusion with estimateAffine2D tests.

* added perf test for estimateAffine functions

tests both estimateAffine2D and estimateAffinePartial2D

* calib3d: compare error in square in estimateAffine2D

* incorporates fix from #6768

* rerun affine estimation on inliers

* stitching: new API for parallel feature finding

due to ABI breakage new functionality is added to `FeaturesFinder2`, `SurfFeaturesFinder2` and `OrbFeaturesFinder2`

* stitching: add tests for parallel feature find API

* perf test (about linear speed up)
* accuracy test compares results with serial version

* stitching: use dynamic_cast to overcome ABI issues

adding parallel API to FeaturesFinder breaks ABI. This commit uses dynamic_cast and hardcodes thread-safe finders to avoid breaking ABI.

This should be replaced by proper method similar to FeaturesMatcher on next ABI break.

* use estimateAffinePartial2D in AffineBestOf2NearestMatcher

* add constructor to AffineBestOf2NearestMatcher

* allows to choose between full affine transform and partial affine transform. Other params are the as for BestOf2NearestMatcher
* added protected field

* samples: stitching_detailed support affine estimator and matcher

* added new flags to choose matcher and estimator

* stitching: rework affine matcher

represent transformation in homogeneous coordinates

affine matcher: remove duplicite code
rework flow to get rid of duplicite code

affine matcher: do not center points to (0, 0)
it is not needed for affine model. it should not affect estimation in any way.

affine matcher: remove unneeded cv namespacing

* stitching: add stub bundle adjuster

* adds stub bundle adjuster that does nothing
* can be used in place of standard bundle adjusters to omit bundle adjusting step

* samples: stitching detailed, support no budle adjust

* uses new NoBundleAdjuster

* added affine warper

* uses R to get whole affine transformation and propagates rotation and translation to plane warper

* add affine warper factory class

* affine warper: compensate transformation

* samples: stitching_detailed add support for affine warper

* add Stitcher::create method

this method follows similar constructor methods and returns smart pointer. This allows constructing Stitcher according to OpenCV guidelines.

* supports multiple stitcher configurations (PANORAMA and SCANS) for convenient setup
* returns cv::Ptr

* stitcher: dynamicaly determine correct estimator

we need to use affine estimator for affine matcher

* preserves ABI (but add hints for ABI 4)
* uses dynamic_cast hack to inject correct estimator

* sample stitching: add support for multiple modes

shows how to use different configurations of stitcher easily (panorama stitching and scans affine model)

* stitcher: find features in parallel

use new FeatureFinder API to find features in parallel. Parallelized using TBB.

* stitching: disable parallel feature finding for OCL

it does not bring much speedup to run features finder in parallel when OpenCL is enabled, because finder needs to wait for OCL device.

Also, currently ORB is not thread-safe when OCL is enabled.

* stitching: move matcher tests

move matchers tests perf_stich.cpp -> perf_matchers.cpp

* stitching: add affine stiching integration test

test basic affine stitching (SCANS mode of stitcher) with images that have only translation between them

* enable surf for stitching tests

stitching.b12 test was failing with surf

investigated the issue, surf is producing good result. Transformation is only slightly different from ORB, so that resulting pano does not exactly match ORB's result. That caused sanity check to fail.

* added size checks similar to other tests
* sanity check will be applied only for ORB

* stitching: fix wrong estimator choice

if case was exactly wrong, estimators were chosen wrong

added logging for estimated transformation

* enable surf for matchers stitching tests

* enable SURF
* rework sanity checking. Check estimated transform instead of matches. Est. transform should be more stable and comparable between SURF and ORB.
* remove regression checking for VectorFeatures tests. It has a lot if data andtest is the same as previous except it test different vector size for performance, so sanity checking does not add any value here. Added basic sanity asserts instead.

* stitching tests: allow relative error for transform

* allows .01 relative error for estimated homography sanity check in stitching matchers tests
* fix VS warning

stitching tests: increase relative error

increase relative error to make it pass on all platforms (results are still good).

stitching test: allow bigger relative error

transformation can differ in small values (with small absolute difference, but large relative difference). transformation output still looks usable for all platforms. This difference affects only mac and windows, linux passes fine with small difference.

* stitching: add tests for affine matcher

uses s1, s2 images. added also new sanity data.

* stitching tests: use different data for matchers tests

this data should yeild more stable transformation (it has much more matches, especially for surf). Sanity data regenerated.

* stitching test: rework tests for matchers

* separated rotation and translations as they are different by scale.
* use appropriate absolute error for them separately. (relative error does not work for values near zero.)

* stitching: fix affine warper compensation

calculation of rotation and translation extracted for plane warper was wrong

* stitching test: enable surf for opencl integration tests

* enable SURF with correct guard (HAVE_OPENCV_XFEATURES2D)
* add OPENCL guard and correct namespace as usual for opencl tests

* stitching: add ocl accuracy test for affine warper

test consistent results with ocl on and off

* stitching: add affine warper ocl perf test

add affine warper to existing warper perf tests. Added new sanity data.

* stitching: do not overwrite inliers in affine matcher

* estimation is run second time on inliers only, inliers produces in second run will not be therefore correct for all matches

* calib3d: add Levenberg–Marquardt refining to estimateAffine2D* functions

this adds affine Levenberg–Marquardt refining to estimateAffine2D functions similar to what is done in findHomography.

implements Levenberg–Marquardt refinig for both full affine and partial affine transformations.

* stitching: remove reestimation step in affine matcher

reestimation step is not needed. estimateAffine2D* functions are running their own reestimation on inliers using the Levenberg-Marquardt algorithm, which is better than simply rerunning RANSAC on inliers.

* implement partial affine bundle adjuster

bundle adjuster that expect affine transform with 4DOF. Refines parameters for all cameras together.

stitching: fix bug in BundleAdjusterAffinePartial

* use the invers properly
* use static buffer for invers to speed it up

* samples: add affine bundle adjuster option to stitching_detailed

* add support for using affine bundle adjuster with 4DOF
* improve logging of initial intristics

* sttiching: add affine bundle adjuster test

* fix build warnings

* stitching: increase limit on sanity check

prevents spurious test failures on mac. values are still pretty fine.

* stitching: set affine bundle adjuster for SCANS mode

* fix bug with AffineBestOf2NearestMatcher (we want to select affine partial mode)
* select right bundle adjuster

* stitching: increase error bound for matcher tests

* this prevents failure on mac. tranformation is still ok.

* stitching: implement affine bundle adjuster

* implements affine bundle adjuster that is using full affine transform
* existing test case modified to test both affinePartial an full affine bundle adjuster

* add stitching tutorial

* show basic usage of stitching api (Stitcher class)

* stitching: add more integration test for affine stitching

* added new datasets to existing testcase
* removed unused include

* calib3d: move `haveCollinearPoints` to common header

* added comment to make that this also checks too close points

* calib3d: redone checkSubset for estimateAffine* callback

* use common function to check collinearity
* this also ensures that point will not be too close to each other

* calib3d: change estimateAffine* functions API

* more similar to `findHomography`, `findFundamentalMat`, `findEssentialMat` and similar
* follows standard recommended semantic INPUTS, OUTPUTS, FLAGS
* allows to disable refining
* supported LMEDS robust method (tests yet to come) along with RANSAC
* extended docs with some tips

* calib3d: rewrite estimateAffine2D test

* rewrite in googletest style
* parametrize to test both robust methods (RANSAC and LMEDS)
* get rid of boilerplate

* calib3d: rework estimateAffinePartial2D test

* rework in googletest style
* add testing for LMEDS

* calib3d: rework estimateAffine*2D perf test

* test for LMEDS speed
* test with/without Levenberg-Marquart
* remove sanity checking (this is covered by accuracy tests)

* calib3d: improve estimateAffine*2D tests

* test transformations in loop
* improves test by testing more potential transformations

* calib3d: rewrite kernels for estimateAffine*2D functions

* use analytical solution instead of SVD
* this version is faster especially for smaller amount of points

* calib3d: tune up perf of estimateAffine*2D functions

* avoid copying inliers
* avoid converting input points if not necessary
* check only `from` point for collinearity, as `to` does not affect stability of transform

* tutorials: add commands examples to stitching tutorials

* add some examples how to run stitcher sample code
* mention stitching_detailed.cpp

* calib3d: change computeError for estimateAffine*2D

* do error computing in floats instead of doubles

this have required precision + we were storing the result in float anyway. This make code faster and allows auto-vectorization by smart compilers.

* documentation: mention estimateAffine*2D function

* refer to new functions on appropriate places
* prefer estimateAffine*2D over estimateRigidTransform

* stitching: add camera models documentations

* mention camera models in module documentation to give user a better overview and reduce confusion

7 years agoMerge pull request #7518 from ArkadiuszRaj:aravis-buffer-status
Arek [Sat, 22 Oct 2016 16:07:55 +0000 (18:07 +0200)]
Merge pull request #7518 from ArkadiuszRaj:aravis-buffer-status

Aravis several updates

* Fix adressing camera with id=0

* Aravis buffer property control & status added

* Modify of autoexposure algorith, ream frame ID from aravis + new properites

* Change of macro name

* VideoCapture now returns no frame on camera disconnecion

* Allow aravis-0.4 usage, proper camera object release.

7 years agoMerge pull request #7536 from alalek:viz_warn
Alexander Alekhin [Fri, 21 Oct 2016 18:21:50 +0000 (18:21 +0000)]
Merge pull request #7536 from alalek:viz_warn

7 years agoviz: eliminate warnings from tutorial code
Alexander Alekhin [Fri, 21 Oct 2016 15:22:07 +0000 (18:22 +0300)]
viz: eliminate warnings from tutorial code

7 years agoviz: update for VTK6.2+
Alexander Alekhin [Fri, 21 Oct 2016 15:19:15 +0000 (18:19 +0300)]
viz: update for VTK6.2+

PCL PR 1205

7 years agoRevert "Fix for VTK6.2 issue."
Alexander Alekhin [Fri, 21 Oct 2016 14:57:33 +0000 (17:57 +0300)]
Revert "Fix for VTK6.2 issue."

This reverts commit f45da9866a94dbf52f5fb955bea67fce913ca515.