Merge branch 4.x

apps/traincascade/imagestorage.cpp

@@ -54,7 +54,7 @@ bool CvCascadeImageReader::NegReader::nextImg()
     size_t count = imgFilenames.size();
     for( size_t i = 0; i < count; i++ )
     {
-        src = imread( imgFilenames[last++], 0 );
+        src = imread( imgFilenames[last++], IMREAD_GRAYSCALE );
         if( src.empty() ){
             last %= count;
             continue;

doc/opencv.bib

@@ -1376,3 +1376,10 @@
   journal = {IEEE transactions on pattern analysis and machine intelligence},
   doi = {10.1109/TPAMI.2006.153}
 }
+@article{Buades2005DenoisingIS,
+  title={Denoising image sequences does not require motion estimation},
+  author={Antoni Buades and Bartomeu Coll and Jean-Michel Morel},
+  journal={IEEE Conference on Advanced Video and Signal Based Surveillance, 2005.},
+  year={2005},
+  pages={70-74}
+}

doc/py_tutorials/py_calib3d/py_depthmap/py_depthmap.markdown

@@ -41,8 +41,8 @@ import numpy as np
 import cv2 as cv
 from matplotlib import pyplot as plt
 
-imgL = cv.imread('tsukuba_l.png',0)
-imgR = cv.imread('tsukuba_r.png',0)
+imgL = cv.imread('tsukuba_l.png', cv.IMREAD_GRAYSCALE)
+imgR = cv.imread('tsukuba_r.png', cv.IMREAD_GRAYSCALE)
 
 stereo = cv.StereoBM_create(numDisparities=16, blockSize=15)
 disparity = stereo.compute(imgL,imgR)

doc/py_tutorials/py_calib3d/py_epipolar_geometry/py_epipolar_geometry.markdown

@@ -76,8 +76,8 @@ import numpy as np
 import cv2 as cv
 from matplotlib import pyplot as plt
 
-img1 = cv.imread('myleft.jpg',0)  #queryimage # left image
-img2 = cv.imread('myright.jpg',0) #trainimage # right image
+img1 = cv.imread('myleft.jpg', cv.IMREAD_GRAYSCALE)  #queryimage # left image
+img2 = cv.imread('myright.jpg', cv.IMREAD_GRAYSCALE) #trainimage # right image
 
 sift = cv.SIFT_create()
 

doc/py_tutorials/py_core/py_basic_ops/py_basic_ops.markdown

@@ -25,6 +25,7 @@ Let's load a color image first:
 >>> import cv2 as cv
 
 >>> img = cv.imread('messi5.jpg')
+>>> assert img is not None, "file could not be read, check with os.path.exists()"
 @endcode
 You can access a pixel value by its row and column coordinates. For BGR image, it returns an array
 of Blue, Green, Red values. For grayscale image, just corresponding intensity is returned.
@@ -173,6 +174,7 @@ from matplotlib import pyplot as plt
 BLUE = [255,0,0]
 
 img1 = cv.imread('opencv-logo.png')
+assert img1 is not None, "file could not be read, check with os.path.exists()"
 
 replicate = cv.copyMakeBorder(img1,10,10,10,10,cv.BORDER_REPLICATE)
 reflect = cv.copyMakeBorder(img1,10,10,10,10,cv.BORDER_REFLECT)

doc/py_tutorials/py_core/py_image_arithmetics/py_image_arithmetics.markdown

@@ -50,6 +50,8 @@ Here \f$\gamma\f$ is taken as zero.
 @code{.py}
 img1 = cv.imread('ml.png')
 img2 = cv.imread('opencv-logo.png')
+assert img1 is not None, "file could not be read, check with os.path.exists()"
+assert img2 is not None, "file could not be read, check with os.path.exists()"
 
 dst = cv.addWeighted(img1,0.7,img2,0.3,0)
 
@@ -76,6 +78,8 @@ bitwise operations as shown below:
 # Load two images
 img1 = cv.imread('messi5.jpg')
 img2 = cv.imread('opencv-logo-white.png')
+assert img1 is not None, "file could not be read, check with os.path.exists()"
+assert img2 is not None, "file could not be read, check with os.path.exists()"
 
 # I want to put logo on top-left corner, So I create a ROI
 rows,cols,channels = img2.shape

doc/py_tutorials/py_core/py_optimization/py_optimization.markdown

@@ -37,6 +37,7 @@ of odd sizes ranging from 5 to 49. (Don't worry about what the result will look
 goal):
 @code{.py}
 img1 = cv.imread('messi5.jpg')
+assert img1 is not None, "file could not be read, check with os.path.exists()"
 
 e1 = cv.getTickCount()
 for i in range(5,49,2):

doc/py_tutorials/py_feature2d/py_brief/py_brief.markdown

@@ -63,7 +63,7 @@ import numpy as np
 import cv2 as cv
 from matplotlib import pyplot as plt
 
-img = cv.imread('simple.jpg',0)
+img = cv.imread('simple.jpg', cv.IMREAD_GRAYSCALE)
 
 # Initiate FAST detector
 star = cv.xfeatures2d.StarDetector_create()

doc/py_tutorials/py_feature2d/py_fast/py_fast.markdown

@@ -98,7 +98,7 @@ import numpy as np
 import cv2 as cv
 from matplotlib import pyplot as plt
 
-img = cv.imread('blox.jpg',0) # `<opencv_root>/samples/data/blox.jpg`
+img = cv.imread('blox.jpg', cv.IMREAD_GRAYSCALE) # `<opencv_root>/samples/data/blox.jpg`
 
 # Initiate FAST object with default values
 fast = cv.FastFeatureDetector_create()

doc/py_tutorials/py_feature2d/py_feature_homography/py_feature_homography.markdown

@@ -40,8 +40,8 @@ from matplotlib import pyplot as plt
 
 MIN_MATCH_COUNT = 10
 
-img1 = cv.imread('box.png',0)          # queryImage
-img2 = cv.imread('box_in_scene.png',0) # trainImage
+img1 = cv.imread('box.png', cv.IMREAD_GRAYSCALE)          # queryImage
+img2 = cv.imread('box_in_scene.png', cv.IMREAD_GRAYSCALE) # trainImage
 
 # Initiate SIFT detector
 sift = cv.SIFT_create()

doc/py_tutorials/py_feature2d/py_orb/py_orb.markdown

@@ -67,7 +67,7 @@ import numpy as np
 import cv2 as cv
 from matplotlib import pyplot as plt
 
-img = cv.imread('simple.jpg',0)
+img = cv.imread('simple.jpg', cv.IMREAD_GRAYSCALE)
 
 # Initiate ORB detector
 orb = cv.ORB_create()

doc/py_tutorials/py_feature2d/py_surf_intro/py_surf_intro.markdown

@@ -76,7 +76,7 @@ and descriptors.
 First we will see a simple demo on how to find SURF keypoints and descriptors and draw it. All
 examples are shown in Python terminal since it is just same as SIFT only.
 @code{.py}
->>> img = cv.imread('fly.png',0)
+>>> img = cv.imread('fly.png', cv.IMREAD_GRAYSCALE)
 
 # Create SURF object. You can specify params here or later.
 # Here I set Hessian Threshold to 400

doc/py_tutorials/py_imgproc/py_canny/py_canny.markdown

@@ -83,7 +83,8 @@ import numpy as np
 import cv2 as cv
 from matplotlib import pyplot as plt
 
-img = cv.imread('messi5.jpg',0)
+img = cv.imread('messi5.jpg', cv.IMREAD_GRAYSCALE)
+assert img is not None, "file could not be read, check with os.path.exists()"
 edges = cv.Canny(img,100,200)
 
 plt.subplot(121),plt.imshow(img,cmap = 'gray')

doc/py_tutorials/py_imgproc/py_contours/py_contour_features/py_contour_features.markdown

@@ -24,7 +24,8 @@ The function **cv.moments()** gives a dictionary of all moment values calculated
 import numpy as np
 import cv2 as cv
 
-img = cv.imread('star.jpg',0)
+img = cv.imread('star.jpg', cv.IMREAD_GRAYSCALE)
+assert img is not None, "file could not be read, check with os.path.exists()"
 ret,thresh = cv.threshold(img,127,255,0)
 contours,hierarchy = cv.findContours(thresh, 1, 2)
 

doc/py_tutorials/py_imgproc/py_contours/py_contours_begin/py_contours_begin.markdown

@@ -29,6 +29,7 @@ import numpy as np
 import cv2 as cv
 
 im = cv.imread('test.jpg')
+assert im is not None, "file could not be read, check with os.path.exists()"
 imgray = cv.cvtColor(im, cv.COLOR_BGR2GRAY)
 ret, thresh = cv.threshold(imgray, 127, 255, 0)
 contours, hierarchy = cv.findContours(thresh, cv.RETR_TREE, cv.CHAIN_APPROX_SIMPLE)

doc/py_tutorials/py_imgproc/py_contours/py_contours_more_functions/py_contours_more_functions.markdown

@@ -41,6 +41,7 @@ import cv2 as cv
 import numpy as np
 
 img = cv.imread('star.jpg')
+assert img is not None, "file could not be read, check with os.path.exists()"
 img_gray = cv.cvtColor(img,cv.COLOR_BGR2GRAY)
 ret,thresh = cv.threshold(img_gray, 127, 255,0)
 contours,hierarchy = cv.findContours(thresh,2,1)
@@ -92,8 +93,10 @@ docs.
 import cv2 as cv
 import numpy as np
 
-img1 = cv.imread('star.jpg',0)
-img2 = cv.imread('star2.jpg',0)
+img1 = cv.imread('star.jpg', cv.IMREAD_GRAYSCALE)
+img2 = cv.imread('star2.jpg', cv.IMREAD_GRAYSCALE)
+assert img1 is not None, "file could not be read, check with os.path.exists()"
+assert img2 is not None, "file could not be read, check with os.path.exists()"
 
 ret, thresh = cv.threshold(img1, 127, 255,0)
 ret, thresh2 = cv.threshold(img2, 127, 255,0)

doc/py_tutorials/py_imgproc/py_filtering/py_filtering.markdown

@@ -29,6 +29,7 @@ import cv2 as cv
 from matplotlib import pyplot as plt
 
 img = cv.imread('opencv_logo.png')
+assert img is not None, "file could not be read, check with os.path.exists()"
 
 kernel = np.ones((5,5),np.float32)/25
 dst = cv.filter2D(img,-1,kernel)
@@ -70,6 +71,7 @@ import numpy as np
 from matplotlib import pyplot as plt
 
 img = cv.imread('opencv-logo-white.png')
+assert img is not None, "file could not be read, check with os.path.exists()"
 
 blur = cv.blur(img,(5,5))
 

doc/py_tutorials/py_imgproc/py_geometric_transformations/py_geometric_transformations.markdown

@@ -28,6 +28,7 @@ import numpy as np
 import cv2 as cv
 
 img = cv.imread('messi5.jpg')
+assert img is not None, "file could not be read, check with os.path.exists()"
 
 res = cv.resize(img,None,fx=2, fy=2, interpolation = cv.INTER_CUBIC)
 
@@ -49,7 +50,8 @@ function. See the below example for a shift of (100,50):
 import numpy as np
 import cv2 as cv
 
-img = cv.imread('messi5.jpg',0)
+img = cv.imread('messi5.jpg', cv.IMREAD_GRAYSCALE)
+assert img is not None, "file could not be read, check with os.path.exists()"
 rows,cols = img.shape
 
 M = np.float32([[1,0,100],[0,1,50]])
@@ -87,7 +89,8 @@ where:
 To find this transformation matrix, OpenCV provides a function, **cv.getRotationMatrix2D**. Check out the
 below example which rotates the image by 90 degree with respect to center without any scaling.
 @code{.py}
-img = cv.imread('messi5.jpg',0)
+img = cv.imread('messi5.jpg', cv.IMREAD_GRAYSCALE)
+assert img is not None, "file could not be read, check with os.path.exists()"
 rows,cols = img.shape
 
 # cols-1 and rows-1 are the coordinate limits.
@@ -108,6 +111,7 @@ which is to be passed to **cv.warpAffine**.
 Check the below example, and also look at the points I selected (which are marked in green color):
 @code{.py}
 img = cv.imread('drawing.png')
+assert img is not None, "file could not be read, check with os.path.exists()"
 rows,cols,ch = img.shape
 
 pts1 = np.float32([[50,50],[200,50],[50,200]])
@@ -137,6 +141,7 @@ matrix.
 See the code below:
 @code{.py}
 img = cv.imread('sudoku.png')
+assert img is not None, "file could not be read, check with os.path.exists()"
 rows,cols,ch = img.shape
 
 pts1 = np.float32([[56,65],[368,52],[28,387],[389,390]])

doc/py_tutorials/py_imgproc/py_grabcut/py_grabcut.markdown

@@ -93,6 +93,7 @@ import cv2 as cv
 from matplotlib import pyplot as plt
 
 img = cv.imread('messi5.jpg')
+assert img is not None, "file could not be read, check with os.path.exists()"
 mask = np.zeros(img.shape[:2],np.uint8)
 
 bgdModel = np.zeros((1,65),np.float64)
@@ -122,7 +123,8 @@ remaining background with gray. Then loaded that mask image in OpenCV, edited or
 got with corresponding values in newly added mask image. Check the code below:*
 @code{.py}
 # newmask is the mask image I manually labelled
-newmask = cv.imread('newmask.png',0)
+newmask = cv.imread('newmask.png', cv.IMREAD_GRAYSCALE)
+assert newmask is not None, "file could not be read, check with os.path.exists()"
 
 # wherever it is marked white (sure foreground), change mask=1
 # wherever it is marked black (sure background), change mask=0

doc/py_tutorials/py_imgproc/py_gradients/py_gradients.markdown

@@ -42,7 +42,8 @@ import numpy as np
 import cv2 as cv
 from matplotlib import pyplot as plt
 
-img = cv.imread('dave.jpg',0)
+img = cv.imread('dave.jpg', cv.IMREAD_GRAYSCALE)
+assert img is not None, "file could not be read, check with os.path.exists()"
 
 laplacian = cv.Laplacian(img,cv.CV_64F)
 sobelx = cv.Sobel(img,cv.CV_64F,1,0,ksize=5)
@@ -79,7 +80,8 @@ import numpy as np
 import cv2 as cv
 from matplotlib import pyplot as plt
 
-img = cv.imread('box.png',0)
+img = cv.imread('box.png', cv.IMREAD_GRAYSCALE)
+assert img is not None, "file could not be read, check with os.path.exists()"
 
 # Output dtype = cv.CV_8U
 sobelx8u = cv.Sobel(img,cv.CV_8U,1,0,ksize=5)

doc/py_tutorials/py_imgproc/py_histograms/py_2d_histogram/py_2d_histogram.markdown

@@ -38,6 +38,7 @@ import numpy as np
 import cv2 as cv
 
 img = cv.imread('home.jpg')
+assert img is not None, "file could not be read, check with os.path.exists()"
 hsv = cv.cvtColor(img,cv.COLOR_BGR2HSV)
 
 hist = cv.calcHist([hsv], [0, 1], None, [180, 256], [0, 180, 0, 256])
@@ -55,6 +56,7 @@ import cv2 as cv
 from matplotlib import pyplot as plt
 
 img = cv.imread('home.jpg')
+assert img is not None, "file could not be read, check with os.path.exists()"
 hsv = cv.cvtColor(img,cv.COLOR_BGR2HSV)
 
 hist, xbins, ybins = np.histogram2d(h.ravel(),s.ravel(),[180,256],[[0,180],[0,256]])
@@ -89,6 +91,7 @@ import cv2 as cv
 from matplotlib import pyplot as plt
 
 img = cv.imread('home.jpg')
+assert img is not None, "file could not be read, check with os.path.exists()"
 hsv = cv.cvtColor(img,cv.COLOR_BGR2HSV)
 hist = cv.calcHist( [hsv], [0, 1], None, [180, 256], [0, 180, 0, 256] )
 

doc/py_tutorials/py_imgproc/py_histograms/py_histogram_backprojection/py_histogram_backprojection.markdown

@@ -38,10 +38,12 @@ import cv2 as cvfrom matplotlib import pyplot as plt
 
 #roi is the object or region of object we need to find
 roi = cv.imread('rose_red.png')
+assert roi is not None, "file could not be read, check with os.path.exists()"
 hsv = cv.cvtColor(roi,cv.COLOR_BGR2HSV)
 
 #target is the image we search in
 target = cv.imread('rose.png')
+assert target is not None, "file could not be read, check with os.path.exists()"
 hsvt = cv.cvtColor(target,cv.COLOR_BGR2HSV)
 
 # Find the histograms using calcHist. Can be done with np.histogram2d also
@@ -85,9 +87,11 @@ import numpy as np
 import cv2 as cv
 
 roi = cv.imread('rose_red.png')
+assert roi is not None, "file could not be read, check with os.path.exists()"
 hsv = cv.cvtColor(roi,cv.COLOR_BGR2HSV)
 
 target = cv.imread('rose.png')
+assert target is not None, "file could not be read, check with os.path.exists()"
 hsvt = cv.cvtColor(target,cv.COLOR_BGR2HSV)
 
 # calculating object histogram

doc/py_tutorials/py_imgproc/py_histograms/py_histogram_begins/py_histogram_begins.markdown

@@ -77,7 +77,8 @@ and its parameters :
 So let's start with a sample image. Simply load an image in grayscale mode and find its full
 histogram.
 @code{.py}
-img = cv.imread('home.jpg',0)
+img = cv.imread('home.jpg', cv.IMREAD_GRAYSCALE)
+assert img is not None, "file could not be read, check with os.path.exists()"
 hist = cv.calcHist([img],[0],None,[256],[0,256])
 @endcode
 hist is a 256x1 array, each value corresponds to number of pixels in that image with its
@@ -121,7 +122,8 @@ import numpy as np
 import cv2 as cv
 from matplotlib import pyplot as plt
 
-img = cv.imread('home.jpg',0)
+img = cv.imread('home.jpg', cv.IMREAD_GRAYSCALE)
+assert img is not None, "file could not be read, check with os.path.exists()"
 plt.hist(img.ravel(),256,[0,256]); plt.show()
 @endcode
 You will get a plot as below :
@@ -136,6 +138,7 @@ import cv2 as cv
 from matplotlib import pyplot as plt
 
 img = cv.imread('home.jpg')
+assert img is not None, "file could not be read, check with os.path.exists()"
 color = ('b','g','r')
 for i,col in enumerate(color):
     histr = cv.calcHist([img],[i],None,[256],[0,256])
@@ -164,7 +167,8 @@ We used cv.calcHist() to find the histogram of the full image. What if you want
 of some regions of an image? Just create a mask image with white color on the region you want to
 find histogram and black otherwise. Then pass this as the mask.
 @code{.py}
-img = cv.imread('home.jpg',0)
+img = cv.imread('home.jpg', cv.IMREAD_GRAYSCALE)
+assert img is not None, "file could not be read, check with os.path.exists()"
 
 # create a mask
 mask = np.zeros(img.shape[:2], np.uint8)

doc/py_tutorials/py_imgproc/py_histograms/py_histogram_equalization/py_histogram_equalization.markdown

@@ -30,7 +30,8 @@ import numpy as np
 import cv2 as cv
 from matplotlib import pyplot as plt
 
-img = cv.imread('wiki.jpg',0)
+img = cv.imread('wiki.jpg', cv.IMREAD_GRAYSCALE)
+assert img is not None, "file could not be read, check with os.path.exists()"
 
 hist,bins = np.histogram(img.flatten(),256,[0,256])
 
@@ -81,7 +82,8 @@ output is our histogram equalized image.
 
 Below is a simple code snippet showing its usage for same image we used :
 @code{.py}
-img = cv.imread('wiki.jpg',0)
+img = cv.imread('wiki.jpg', cv.IMREAD_GRAYSCALE)
+assert img is not None, "file could not be read, check with os.path.exists()"
 equ = cv.equalizeHist(img)
 res = np.hstack((img,equ)) #stacking images side-by-side
 cv.imwrite('res.png',res)
@@ -124,7 +126,8 @@ Below code snippet shows how to apply CLAHE in OpenCV:
 import numpy as np
 import cv2 as cv
 
-img = cv.imread('tsukuba_l.png',0)
+img = cv.imread('tsukuba_l.png', cv.IMREAD_GRAYSCALE)
+assert img is not None, "file could not be read, check with os.path.exists()"
 
 # create a CLAHE object (Arguments are optional).
 clahe = cv.createCLAHE(clipLimit=2.0, tileGridSize=(8,8))

doc/py_tutorials/py_imgproc/py_houghcircles/py_houghcircles.markdown

@@ -23,7 +23,8 @@ explained in the documentation. So we directly go to the code.
 import numpy as np
 import cv2 as cv
 
-img = cv.imread('opencv-logo-white.png',0)
+img = cv.imread('opencv-logo-white.png', cv.IMREAD_GRAYSCALE)
+assert img is not None, "file could not be read, check with os.path.exists()"
 img = cv.medianBlur(img,5)
 cimg = cv.cvtColor(img,cv.COLOR_GRAY2BGR)
 

doc/py_tutorials/py_imgproc/py_morphological_ops/py_morphological_ops.markdown

@@ -38,7 +38,8 @@ Here, as an example, I would use a 5x5 kernel with full of ones. Let's see it ho
 import cv2 as cv
 import numpy as np
 
-img = cv.imread('j.png',0)
+img = cv.imread('j.png', cv.IMREAD_GRAYSCALE)
+assert img is not None, "file could not be read, check with os.path.exists()"
 kernel = np.ones((5,5),np.uint8)
 erosion = cv.erode(img,kernel,iterations = 1)
 @endcode

doc/py_tutorials/py_imgproc/py_pyramids/py_pyramids.markdown

@@ -31,6 +31,7 @@ Similarly while expanding, area becomes 4 times in each level. We can find Gauss
 **cv.pyrDown()** and **cv.pyrUp()** functions.
 @code{.py}
 img = cv.imread('messi5.jpg')
+assert img is not None, "file could not be read, check with os.path.exists()"
 lower_reso = cv.pyrDown(higher_reso)
 @endcode
 Below is the 4 levels in an image pyramid.
@@ -84,6 +85,8 @@ import numpy as np,sys
 
 A = cv.imread('apple.jpg')
 B = cv.imread('orange.jpg')
+assert A is not None, "file could not be read, check with os.path.exists()"
+assert B is not None, "file could not be read, check with os.path.exists()"
 
 # generate Gaussian pyramid for A
 G = A.copy()

doc/py_tutorials/py_imgproc/py_template_matching/py_template_matching.markdown

@@ -38,9 +38,11 @@ import cv2 as cv
 import numpy as np
 from matplotlib import pyplot as plt
 
-img = cv.imread('messi5.jpg',0)
+img = cv.imread('messi5.jpg', cv.IMREAD_GRAYSCALE)
+assert img is not None, "file could not be read, check with os.path.exists()"
 img2 = img.copy()
-template = cv.imread('template.jpg',0)
+template = cv.imread('template.jpg', cv.IMREAD_GRAYSCALE)
+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
@@ -113,8 +115,10 @@ import numpy as np
 from matplotlib import pyplot as plt
 
 img_rgb = cv.imread('mario.png')
+assert img_rgb is not None, "file could not be read, check with os.path.exists()"
 img_gray = cv.cvtColor(img_rgb, cv.COLOR_BGR2GRAY)
-template = cv.imread('mario_coin.png',0)
+template = cv.imread('mario_coin.png', cv.IMREAD_GRAYSCALE)
+assert template is not None, "file could not be read, check with os.path.exists()"
 w, h = template.shape[::-1]
 
 res = cv.matchTemplate(img_gray,template,cv.TM_CCOEFF_NORMED)

doc/py_tutorials/py_imgproc/py_thresholding/py_thresholding.markdown

@@ -37,7 +37,8 @@ import cv2 as cv
 import numpy as np
 from matplotlib import pyplot as plt
 
-img = cv.imread('gradient.png',0)
+img = cv.imread('gradient.png', cv.IMREAD_GRAYSCALE)
+assert img is not None, "file could not be read, check with os.path.exists()"
 ret,thresh1 = cv.threshold(img,127,255,cv.THRESH_BINARY)
 ret,thresh2 = cv.threshold(img,127,255,cv.THRESH_BINARY_INV)
 ret,thresh3 = cv.threshold(img,127,255,cv.THRESH_TRUNC)
@@ -85,7 +86,8 @@ import cv2 as cv
 import numpy as np
 from matplotlib import pyplot as plt
 
-img = cv.imread('sudoku.png',0)
+img = cv.imread('sudoku.png', cv.IMREAD_GRAYSCALE)
+assert img is not None, "file could not be read, check with os.path.exists()"
 img = cv.medianBlur(img,5)
 
 ret,th1 = cv.threshold(img,127,255,cv.THRESH_BINARY)
@@ -133,7 +135,8 @@ import cv2 as cv
 import numpy as np
 from matplotlib import pyplot as plt
 
-img = cv.imread('noisy2.png',0)
+img = cv.imread('noisy2.png', cv.IMREAD_GRAYSCALE)
+assert img is not None, "file could not be read, check with os.path.exists()"
 
 # global thresholding
 ret1,th1 = cv.threshold(img,127,255,cv.THRESH_BINARY)
@@ -183,7 +186,8 @@ where
 It actually finds a value of t which lies in between two peaks such that variances to both classes
 are minimal. It can be simply implemented in Python as follows:
 @code{.py}
-img = cv.imread('noisy2.png',0)
+img = cv.imread('noisy2.png', cv.IMREAD_GRAYSCALE)
+assert img is not None, "file could not be read, check with os.path.exists()"
 blur = cv.GaussianBlur(img,(5,5),0)
 
 # find normalized_histogram, and its cumulative distribution function

doc/py_tutorials/py_imgproc/py_transforms/py_fourier_transform/py_fourier_transform.markdown

@@ -54,7 +54,8 @@ import cv2 as cv
 import numpy as np
 from matplotlib import pyplot as plt
 
-img = cv.imread('messi5.jpg',0)
+img = cv.imread('messi5.jpg', cv.IMREAD_GRAYSCALE)
+assert img is not None, "file could not be read, check with os.path.exists()"
 f = np.fft.fft2(img)
 fshift = np.fft.fftshift(f)
 magnitude_spectrum = 20*np.log(np.abs(fshift))
@@ -121,7 +122,8 @@ import numpy as np
 import cv2 as cv
 from matplotlib import pyplot as plt
 
-img = cv.imread('messi5.jpg',0)
+img = cv.imread('messi5.jpg', cv.IMREAD_GRAYSCALE)
+assert img is not None, "file could not be read, check with os.path.exists()"
 
 dft = cv.dft(np.float32(img),flags = cv.DFT_COMPLEX_OUTPUT)
 dft_shift = np.fft.fftshift(dft)
@@ -184,7 +186,8 @@ So how do we find this optimal size ? OpenCV provides a function, **cv.getOptima
 this. It is applicable to both **cv.dft()** and **np.fft.fft2()**. Let's check their performance
 using IPython magic command %timeit.
 @code{.py}
-In [16]: img = cv.imread('messi5.jpg',0)
+In [15]: img = cv.imread('messi5.jpg', cv.IMREAD_GRAYSCALE)
+In [16]: assert img is not None, "file could not be read, check with os.path.exists()"
 In [17]: rows,cols = img.shape
 In [18]: print("{} {}".format(rows,cols))
 342 548

doc/py_tutorials/py_imgproc/py_watershed/py_watershed.markdown

@@ -49,6 +49,7 @@ import cv2 as cv
 from matplotlib import pyplot as plt
 
 img = cv.imread('coins.png')
+assert img is not None, "file could not be read, check with os.path.exists()"
 gray = cv.cvtColor(img,cv.COLOR_BGR2GRAY)
 ret, thresh = cv.threshold(gray,0,255,cv.THRESH_BINARY_INV+cv.THRESH_OTSU)
 @endcode

doc/py_tutorials/py_photo/py_inpainting/py_inpainting.markdown

@@ -56,7 +56,7 @@ import numpy as np
 import cv2 as cv
 
 img = cv.imread('messi_2.jpg')
-mask = cv.imread('mask2.png',0)
+mask = cv.imread('mask2.png', cv.IMREAD_GRAYSCALE)
 
 dst = cv.inpaint(img,mask,3,cv.INPAINT_TELEA)
 

doc/tutorials/introduction/linux_eclipse/linux_eclipse.markdown

@@ -63,7 +63,7 @@ Making a project
     int main( int argc, char** argv )
     {
       Mat image;
-      image = imread( argv[1], 1 );
+      image = imread( argv[1], IMREAD_COLOR );
 
       if( argc != 2 || !image.data )
         {

doc/tutorials/introduction/linux_gcc_cmake/linux_gcc_cmake.markdown

@@ -42,7 +42,7 @@ int main(int argc, char** argv )
     }
 
     Mat image;
-    image = imread( argv[1], 1 );
+    image = imread( argv[1], IMREAD_COLOR );
 
     if ( !image.data )
     {

modules/3d/src/usac/local_optimization.cpp

@@ -20,7 +20,7 @@ protected:
 
     std::vector<int> labeling_inliers;
     std::vector<double> energies, weights;
-    std::vector<bool> used_edges;
+    std::set<int> used_edges;
     std::vector<Mat> gc_models;
 public:
 
@@ -40,7 +40,7 @@ public:
 
         energies = std::vector<double>(points_size);
         labeling_inliers = std::vector<int>(points_size);
-        used_edges = std::vector<bool>(points_size*points_size);
+        used_edges = std::set<int>();
         gc_models = std::vector<Mat> (estimator->getMaxNumSolutionsNonMinimal());
     }
 
@@ -115,7 +115,7 @@ private:
             energies[pt] = energy > 1 ? 1 : energy;
         }
 
-        std::fill(used_edges.begin(), used_edges.end(), false);
+        used_edges.clear();
 
         bool has_edges = false;
         // Iterate through all points and set their edges
@@ -125,12 +125,12 @@ private:
             // Iterate through  all neighbors
             for (int actual_neighbor_idx : neighborhood_graph->getNeighbors(point_idx)) {
                 if (actual_neighbor_idx == point_idx ||
-                    used_edges[actual_neighbor_idx*points_size + point_idx] ||
-                    used_edges[point_idx*points_size + actual_neighbor_idx])
+                    used_edges.count(actual_neighbor_idx*points_size + point_idx) > 0 ||
+                    used_edges.count(point_idx*points_size + actual_neighbor_idx) > 0)
                     continue;
 
-                used_edges[actual_neighbor_idx*points_size + point_idx] = true;
-                used_edges[point_idx*points_size + actual_neighbor_idx] = true;
+                used_edges.insert(actual_neighbor_idx*points_size + point_idx);
+                used_edges.insert(point_idx*points_size + actual_neighbor_idx);
 
                 double a = (0.5 * (energy + energies[actual_neighbor_idx])) * spatial_coherence,
                        b = spatial_coherence, c = spatial_coherence, d = 0;

modules/3d/test/test_homography.cpp

@@ -74,60 +74,27 @@ int METHOD[METHODS_COUNT] = {0, cv::RANSAC, cv::LMEDS, cv::RHO};
 using namespace cv;
 using namespace std;
 
-class CV_HomographyTest: public cvtest::ArrayTest
-{
-public:
-    CV_HomographyTest();
-    ~CV_HomographyTest();
 
-    void run (int);
+namespace HomographyTestUtils {
 
-protected:
+static const float max_diff = 0.032f;
+static const float max_2diff = 0.020f;
+static const int image_size = 100;
+static const double reproj_threshold = 3.0;
+static const double sigma = 0.01;
 
-    int method;
-    int image_size;
-    double reproj_threshold;
-    double sigma;
-
-private:
-    float max_diff, max_2diff;
-    bool check_matrix_size(const cv::Mat& H);
-    bool check_matrix_diff(const cv::Mat& original, const cv::Mat& found, const int norm_type, double &diff);
-    int check_ransac_mask_1(const Mat& src, const Mat& mask);
-    int check_ransac_mask_2(const Mat& original_mask, const Mat& found_mask);
-
-    void print_information_1(int j, int N, int method, const Mat& H);
-    void print_information_2(int j, int N, int method, const Mat& H, const Mat& H_res, int k, double diff);
-    void print_information_3(int method, int j, int N, const Mat& mask);
-    void print_information_4(int method, int j, int N, int k, int l, double diff);
-    void print_information_5(int method, int j, int N, int l, double diff);
-    void print_information_6(int method, int j, int N, int k, double diff, bool value);
-    void print_information_7(int method, int j, int N, int k, double diff, bool original_value, bool found_value);
-    void print_information_8(int method, int j, int N, int k, int l, double diff);
-};
-
-CV_HomographyTest::CV_HomographyTest() : max_diff(1e-2f), max_2diff(2e-2f)
-{
-    method = 0;
-    image_size = 100;
-    reproj_threshold = 3.0;
-    sigma = 0.01;
-}
-
-CV_HomographyTest::~CV_HomographyTest() {}
-
-bool CV_HomographyTest::check_matrix_size(const cv::Mat& H)
+static bool check_matrix_size(const cv::Mat& H)
 {
     return (H.rows == 3) && (H.cols == 3);
 }
 
-bool CV_HomographyTest::check_matrix_diff(const cv::Mat& original, const cv::Mat& found, const int norm_type, double &diff)
+static bool check_matrix_diff(const cv::Mat& original, const cv::Mat& found, const int norm_type, double &diff)
 {
     diff = cvtest::norm(original, found, norm_type);
     return diff <= max_diff;
 }
 
-int CV_HomographyTest::check_ransac_mask_1(const Mat& src, const Mat& mask)
+static int check_ransac_mask_1(const Mat& src, const Mat& mask)
 {
     if (!(mask.cols == 1) && (mask.rows == src.cols)) return 1;
     if (countNonZero(mask) < mask.rows) return 2;
@@ -135,14 +102,14 @@ int CV_HomographyTest::check_ransac_mask_1(const Mat& src, const Mat& mask)
     return 0;
 }
 
-int CV_HomographyTest::check_ransac_mask_2(const Mat& original_mask, const Mat& found_mask)
+static int check_ransac_mask_2(const Mat& original_mask, const Mat& found_mask)
 {
     if (!(found_mask.cols == 1) && (found_mask.rows == original_mask.rows)) return 1;
     for (int i = 0; i < found_mask.rows; ++i) if (found_mask.at<uchar>(i, 0) > 1) return 2;
     return 0;
 }
 
-void CV_HomographyTest::print_information_1(int j, int N, int _method, const Mat& H)
+static void print_information_1(int j, int N, int _method, const Mat& H)
 {
     cout << endl; cout << "Checking for homography matrix sizes..." << endl; cout << endl;
     cout << "Type of srcPoints: "; if ((j>-1) && (j<2)) cout << "Mat of CV_32FC2"; else  cout << "vector <Point2f>";
@@ -154,7 +121,7 @@ void CV_HomographyTest::print_information_1(int j, int N, int _method, const Mat
     cout << "Number of rows: " << H.rows << "   Number of cols: " << H.cols << endl; cout << endl;
 }
 
-void CV_HomographyTest::print_information_2(int j, int N, int _method, const Mat& H, const Mat& H_res, int k, double diff)
+static void print_information_2(int j, int N, int _method, const Mat& H, const Mat& H_res, int k, double diff)
 {
     cout << endl; cout << "Checking for accuracy of homography matrix computing..." << endl; cout << endl;
     cout << "Type of srcPoints: "; if ((j>-1) && (j<2)) cout << "Mat of CV_32FC2"; else  cout << "vector <Point2f>";
@@ -170,7 +137,7 @@ void CV_HomographyTest::print_information_2(int j, int N, int _method, const Mat
     cout << "Maximum allowed difference: " << max_diff << endl; cout << endl;
 }
 
-void CV_HomographyTest::print_information_3(int _method, int j, int N, const Mat& mask)
+static void print_information_3(int _method, int j, int N, const Mat& mask)
 {
     cout << endl; cout << "Checking for inliers/outliers mask..." << endl; cout << endl;
     cout << "Type of srcPoints: "; if ((j>-1) && (j<2)) cout << "Mat of CV_32FC2"; else  cout << "vector <Point2f>";
@@ -182,7 +149,7 @@ void CV_HomographyTest::print_information_3(int _method, int j, int N, const Mat
     cout << "Number of rows: " << mask.rows << "   Number of cols: " << mask.cols << endl; cout << endl;
 }
 
-void CV_HomographyTest::print_information_4(int _method, int j, int N, int k, int l, double diff)
+static void print_information_4(int _method, int j, int N, int k, int l, double diff)
 {
     cout << endl; cout << "Checking for accuracy of reprojection error computing..." << endl; cout << endl;
     cout << "Method: "; if (_method == 0) cout << 0 << endl; else cout << "CV_LMEDS" << endl;
@@ -196,7 +163,7 @@ void CV_HomographyTest::print_information_4(int _method, int j, int N, int k, in
     cout << "Maximum allowed difference: " << max_2diff << endl; cout << endl;
 }
 
-void CV_HomographyTest::print_information_5(int _method, int j, int N, int l, double diff)
+static void print_information_5(int _method, int j, int N, int l, double diff)
 {
     cout << endl; cout << "Checking for accuracy of reprojection error computing..." << endl; cout << endl;
     cout << "Method: "; if (_method == 0) cout << 0 << endl; else cout << "CV_LMEDS" << endl;
@@ -209,7 +176,7 @@ void CV_HomographyTest::print_information_5(int _method, int j, int N, int l, do
     cout << "Maximum allowed difference: " << max_diff << endl; cout << endl;
 }
 
-void CV_HomographyTest::print_information_6(int _method, int j, int N, int k, double diff, bool value)
+static void print_information_6(int _method, int j, int N, int k, double diff, bool value)
 {
     cout << endl; cout << "Checking for inliers/outliers mask..." << endl; cout << endl;
     cout << "Method: "; if (_method == RANSAC) cout << "RANSAC" << endl; else if (_method == cv::RHO) cout << "RHO" << endl; else cout << _method << endl;
@@ -222,7 +189,7 @@ void CV_HomographyTest::print_information_6(int _method, int j, int N, int k, do
     cout << "Value of found mask: "<< value << endl; cout << endl;
 }
 
-void CV_HomographyTest::print_information_7(int _method, int j, int N, int k, double diff, bool original_value, bool found_value)
+static void print_information_7(int _method, int j, int N, int k, double diff, bool original_value, bool found_value)
 {
     cout << endl; cout << "Checking for inliers/outliers mask..." << endl; cout << endl;
     cout << "Method: "; if (_method == RANSAC) cout << "RANSAC" << endl; else if (_method == cv::RHO) cout << "RHO" << endl; else cout << _method << endl;
@@ -235,7 +202,7 @@ void CV_HomographyTest::print_information_7(int _method, int j, int N, int k, do
     cout << "Value of original mask: "<< original_value << "   Value of found mask: " << found_value << endl; cout << endl;
 }
 
-void CV_HomographyTest::print_information_8(int _method, int j, int N, int k, int l, double diff)
+static void print_information_8(int _method, int j, int N, int k, int l, double diff)
 {
     cout << endl; cout << "Checking for reprojection error of inlier..." << endl; cout << endl;
     cout << "Method: "; if (_method == RANSAC) cout << "RANSAC" << endl; else if (_method == cv::RHO) cout << "RHO" << endl; else cout << _method << endl;
@@ -249,11 +216,15 @@ void CV_HomographyTest::print_information_8(int _method, int j, int N, int k, in
     cout << "Maximum allowed difference: " << max_2diff << endl; cout << endl;
 }
 
-void CV_HomographyTest::run(int)
+} // HomographyTestUtils::
+
+
+TEST(Calib3d_Homography, accuracy)
 {
+    using namespace HomographyTestUtils;
     for (int N = MIN_COUNT_OF_POINTS; N <= MAX_COUNT_OF_POINTS; ++N)
     {
-        RNG& rng = ts->get_rng();
+        RNG& rng = cv::theRNG();
 
         float *src_data = new float [2*N];
 
@@ -309,7 +280,7 @@ void CV_HomographyTest::run(int)
 
         for (int i = 0; i < METHODS_COUNT; ++i)
         {
-            method = METHOD[i];
+            const int method = METHOD[i];
             switch (method)
             {
             case 0:
@@ -412,7 +383,7 @@ void CV_HomographyTest::run(int)
 
         for (int i = 0; i < METHODS_COUNT; ++i)
         {
-            method = METHOD[i];
+            const int method = METHOD[i];
             switch (method)
             {
             case 0:
@@ -574,8 +545,6 @@ void CV_HomographyTest::run(int)
     }
 }
 
-TEST(Calib3d_Homography, accuracy) { CV_HomographyTest test; test.safe_run(); }
-
 TEST(Calib3d_Homography, EKcase)
 {
     float pt1data[] =

modules/3d/test/test_usac.cpp

@@ -416,7 +416,7 @@ TEST (usac_Affine2D, accuracy) {
 
 TEST(usac_testUsacParams, accuracy) {
     std::vector<int> gt_inliers;
-    const int pts_size = 1500;
+    const int pts_size = 150000;
     cv::RNG &rng = cv::theRNG();
     const cv::UsacParams usac_params = cv::UsacParams();
     cv::Mat pts1, pts2, K1, K2, mask, model, rvec, tvec, R;

modules/calib/src/calibration.cpp

@@ -404,6 +404,12 @@ static double calibrateCameraInternal( const Mat& objectPoints,
         mask[nparams - 1] = 0;
     }
 
+    int nparams_nz = countNonZero(mask);
+
+    if (nparams_nz >= 2 * total)
+        CV_Error_(Error::StsBadArg,
+                  ("There should be less vars to optimize (having %d) than the number of residuals (%d = 2 per point)", nparams_nz, 2 * total));
+
     // 2. initialize extrinsic parameters
     for(int i = 0, pos = 0; i < nimages; i++ )
     {
@@ -572,26 +578,23 @@ static double calibrateCameraInternal( const Mat& objectPoints,
 
     if (!stdDevs.empty())
     {
-        int nparams_nz = countNonZero(mask);
         JtJN.create(nparams_nz, nparams_nz, CV_64F);
 
         subMatrix(JtJ, JtJN, mask, mask);
         completeSymm(JtJN, false);
         //TODO: try DECOMP_CHOLESKY maybe?
         cv::invert(JtJN, JtJinv, DECOMP_EIG);
-        // sigma2 is deviation of the noise
-        // see any papers about variance of the least squares estimator for
-        // detailed description of the variance estimation methods
-        double sigma2 = norm(allErrors, NORM_L2SQR) / (total - nparams_nz);
+        // an explanation of that denominator correction can be found here:
+        // R. Hartley, A. Zisserman, Multiple View Geometry in Computer Vision, 2004, section 5.1.3, page 134
+        // see the discussion for more details: https://github.com/opencv/opencv/pull/22992
+        double sigma2 = norm(allErrors, NORM_L2SQR) / (2 * total - nparams_nz);
         int j = 0;
         for ( int s = 0; s < nparams; s++ )
+        {
+            stdDevs.at<double>(s) = mask[s] ? std::sqrt(JtJinv.at<double>(j, j) * sigma2) : 0.0;
             if( mask[s] )
-            {
-                stdDevs.at<double>(s) = std::sqrt(JtJinv.at<double>(j,j) * sigma2);
                 j++;
-            }
-            else
-                stdDevs.at<double>(s) = 0.;
+        }
     }
 
     // 4. store the results

modules/calib/test/test_calibration_hand_eye.cpp

@@ -426,9 +426,9 @@ public:
         eps_rvec_noise[CALIB_HAND_EYE_ANDREFF] = 1.0e-2;
         eps_rvec_noise[CALIB_HAND_EYE_DANIILIDIS] = 1.0e-2;
 
-        eps_tvec_noise[CALIB_HAND_EYE_TSAI] = 5.0e-2;
-        eps_tvec_noise[CALIB_HAND_EYE_PARK] = 5.0e-2;
-        eps_tvec_noise[CALIB_HAND_EYE_HORAUD] = 5.0e-2;
+        eps_tvec_noise[CALIB_HAND_EYE_TSAI] = 7.0e-2;
+        eps_tvec_noise[CALIB_HAND_EYE_PARK] = 7.0e-2;
+        eps_tvec_noise[CALIB_HAND_EYE_HORAUD] = 7.0e-2;
         if (eyeToHandConfig)
         {
             eps_tvec_noise[CALIB_HAND_EYE_ANDREFF] = 7.0e-2;
@@ -453,7 +453,7 @@ void CV_CalibrateHandEyeTest::run(int)
 {
     ts->set_failed_test_info(cvtest::TS::OK);
 
-    RNG& rng = ts->get_rng();
+    RNG& rng = cv::theRNG();
 
     std::vector<std::vector<double> > vec_rvec_diff(5);
     std::vector<std::vector<double> > vec_tvec_diff(5);

modules/calib/test/test_chesscorners.cpp

@@ -224,7 +224,7 @@ void CV_ChessboardDetectorTest::run_batch( const string& filename )
 
         /* read the image */
         String img_file = board_list[idx * 2];
-        Mat gray = imread( folder + img_file, 0);
+        Mat gray = imread( folder + img_file, IMREAD_GRAYSCALE);
 
         if( gray.empty() )
         {

modules/core/include/opencv2/core/fast_math.hpp

@@ -306,7 +306,7 @@ CV_INLINE int cvIsInf( double value )
 #elif defined(__x86_64__) || defined(_M_X64) || defined(__aarch64__) || defined(_M_ARM64) || defined(__PPC64__) || defined(__loongarch64)
     Cv64suf ieee754;
     ieee754.f = value;
-    return (ieee754.u & 0x7fffffff00000000) ==
+    return (ieee754.u & 0x7fffffffffffffff) ==
                         0x7ff0000000000000;
 #else
     Cv64suf ieee754;

modules/core/src/parallel.cpp

@@ -275,7 +275,9 @@ namespace {
         void recordException(const cv::String& msg)
 #endif
         {
+#ifndef CV_THREAD_SANITIZER
             if (!hasException)
+#endif
             {
                 cv::AutoLock lock(cv::getInitializationMutex());
                 if (!hasException)

modules/core/src/system.cpp

@@ -247,6 +247,7 @@ std::wstring GetTempFileNameWinRT(std::wstring prefix)
 #if defined __MACH__ && defined __APPLE__
 #include <mach/mach.h>
 #include <mach/mach_time.h>
+#include <sys/sysctl.h>
 #endif
 
 #endif
@@ -635,6 +636,14 @@ struct HWFeatures
     #if (defined __ARM_FP  && (((__ARM_FP & 0x2) != 0) && defined __ARM_NEON__))
         have[CV_CPU_FP16] = true;
     #endif
+    #if (defined __ARM_FEATURE_DOTPROD)
+        int has_feat_dotprod = 0;
+        size_t has_feat_dotprod_size = sizeof(has_feat_dotprod);
+        sysctlbyname("hw.optional.arm.FEAT_DotProd", &has_feat_dotprod, &has_feat_dotprod_size, NULL, 0);
+        if (has_feat_dotprod) {
+            have[CV_CPU_NEON_DOTPROD] = true;
+        }
+    #endif
     #elif (defined __clang__)
     #if (defined __ARM_NEON__ || (defined __ARM_NEON && defined __aarch64__))
         have[CV_CPU_NEON] = true;

modules/core/test/test_math.cpp

@@ -3997,6 +3997,13 @@ TEST(Core_FastMath, InlineNaN)
     EXPECT_EQ( cvIsNaN((double) NAN), 1);
     EXPECT_EQ( cvIsNaN((double) -NAN), 1);
     EXPECT_EQ( cvIsNaN(0.0), 0);
+
+    // Regression: check the +/-Inf cases
+    Cv64suf suf;
+    suf.u = 0x7FF0000000000000UL;
+    EXPECT_EQ( cvIsNaN(suf.f), 0);
+    suf.u = 0xFFF0000000000000UL;
+    EXPECT_EQ( cvIsNaN(suf.f), 0);
 }
 
 TEST(Core_FastMath, InlineIsInf)
@@ -4008,6 +4015,13 @@ TEST(Core_FastMath, InlineIsInf)
     EXPECT_EQ( cvIsInf((double) HUGE_VAL), 1);
     EXPECT_EQ( cvIsInf((double) -HUGE_VAL), 1);
     EXPECT_EQ( cvIsInf(0.0), 0);
+
+    // Regression: check the cases of 0x7FF00000xxxxxxxx
+    Cv64suf suf;
+    suf.u = 0x7FF0000000000001UL;
+    EXPECT_EQ( cvIsInf(suf.f), 0);
+    suf.u = 0x7FF0000012345678UL;
+    EXPECT_EQ( cvIsInf(suf.f), 0);
 }
 
 }} // namespace

modules/dnn/include/opencv2/dnn/all_layers.hpp

@@ -1085,6 +1085,16 @@ CV__DNN_INLINE_NS_BEGIN
         static Ptr<TileLayer> create(const LayerParams& params);
     };
 
+    class CV_EXPORTS LayerNormLayer : public Layer
+    {
+    public:
+        bool hasBias;
+        int axis;
+        float epsilon;
+
+        static Ptr<LayerNormLayer> create(const LayerParams& params);
+    };
+
 //! @}
 //! @}
 CV__DNN_INLINE_NS_END

modules/dnn/perf/perf_layer.cpp

@@ -417,6 +417,212 @@ PERF_TEST_P_(Layer_ScatterND, DISABLED_ScatterND_add)
     test_layer({N, C, H , W}, "add");
 }
 
+struct Layer_LayerNorm : public TestBaseWithParam<tuple<Backend, Target> >
+{
+    void test_layer(const std::vector<int>& x_shape)
+    {
+        int backendId = get<0>(GetParam());
+        int targetId = get<1>(GetParam());
+
+        Mat x(x_shape, CV_32FC1);
+        Mat scale(x_shape.back(), 1, CV_32FC1);
+        Mat b(x_shape.back(), 1, CV_32FC1);
+
+        randu(x, 0.f, 1.f);
+        randu(scale, 0.f, 1.f);
+        randu(b, 0.f, 1.f);
+
+
+        Net net;
+        LayerParams lp;
+        lp.type = "LayerNormalization";
+        lp.name = "testLayer";
+        lp.set("axis", 2);
+        lp.set("hasBias", true);
+        int id = net.addLayerToPrev(lp.name, lp.type, lp);
+        net.connect(0, 0, id, 0);
+        net.connect(0, 1, id, 1);
+        net.connect(0, 2, id, 2);
+
+        // warmup
+        {
+            std::vector<String> inpNames(3);
+            inpNames[0] = "x";
+            inpNames[1] = "scale";
+            inpNames[2] = "b";
+            net.setInputsNames(inpNames);
+            net.setInput(x, inpNames[0]);
+            net.setInput(scale, inpNames[1]);
+            net.setInput(b, inpNames[2]);
+
+            net.setPreferableBackend(backendId);
+            net.setPreferableTarget(targetId);
+            Mat out = net.forward();
+        }
+
+        TEST_CYCLE()
+        {
+            Mat res = net.forward();
+        }
+
+        SANITY_CHECK_NOTHING();
+    }
+
+    int N = 1;
+    int H = 50;
+    int W = 768;
+};
+
+PERF_TEST_P_(Layer_LayerNorm, LayerNorm)
+{
+    test_layer({N, H ,W});
+}
+
+struct Layer_LayerNormExpanded : public TestBaseWithParam<tuple<Backend, Target> >
+{
+    void test_layer(const std::vector<int>& x_shape)
+    {
+        int backendId = get<0>(GetParam());
+        int targetId = get<1>(GetParam());
+
+        Mat x(x_shape, CV_32FC1);
+        Mat scale(1, x_shape.back(), CV_32FC1); // transpose to pass shape check
+        Mat b(1, x_shape.back(), CV_32FC1);     // transpose to pass shape check
+
+        randu(x, 0.f, 1.f);
+        randu(scale, 0.f, 1.f);
+        randu(b, 0.f, 1.f);
+
+        // sub graph structure:
+        //   -> ReduceMean ->     -> Pow(2) -> ReduceMean -> Add(epsilon) -> Sqrt ->
+        // x                  Sub                                                    Div -> Mul(scale) -> Add(bias)
+        //   --------------->     ------------------------------------------------->
+
+        Net net;
+
+        LayerParams lp_rm;
+        lp_rm.type = "Reduce";
+        lp_rm.name = "reducemean1";
+        lp_rm.set("reduce", "AVE");
+        std::vector<int> deleteDims(1, x_shape.back());
+        lp_rm.set("deleted_dims", DictValue::arrayInt(&deleteDims[0], deleteDims.size()));
+        std::vector<int> targetDims(x_shape.begin(), x_shape.end());
+        targetDims[x_shape.size() - 1] = 1;
+        lp_rm.set("target_dims", DictValue::arrayInt(&targetDims[0], targetDims.size()));
+        int id_rm = net.addLayerToPrev(lp_rm.name, lp_rm.type, lp_rm);
+        net.connect(0, 0, id_rm, 0);
+
+        LayerParams lp_sub;
+        lp_sub.type = "NaryEltwise";
+        lp_sub.name = "sub1";
+        lp_sub.set("operation", "sub");
+        int id_sub = net.addLayer(lp_sub.name, lp_sub.type, lp_sub);
+        net.connect(0, 0, id_sub, 0);
+        net.connect(id_rm, 0, id_sub, 1);
+
+        Mat pow_const(1, 1, CV_32FC1);
+        pow_const.at<float>(0) = 2.f;
+        LayerParams lp_pow_const;
+        lp_pow_const.type = "Const";
+        lp_pow_const.name = "const1";
+        lp_pow_const.blobs.push_back(pow_const);
+        int id_pow_const = net.addLayer(lp_pow_const.name, lp_pow_const.type, lp_pow_const);
+        LayerParams lp_pow;
+        lp_pow.type = "NaryEltwise";
+        lp_pow.name = "pow1";
+        lp_pow.set("operation", "pow");
+        int id_pow = net.addLayer(lp_pow.name, lp_pow.type, lp_pow);
+        net.connect(id_sub, 0, id_pow, 0);
+        net.connect(id_pow_const, 0, id_pow, 1);
+
+        LayerParams lp_rm1;
+        lp_rm1.type = "Reduce";
+        lp_rm1.name = "reducemean2";
+        lp_rm1.set("reduce", "AVE");
+        lp_rm1.set("deleted_dims", DictValue::arrayInt(&deleteDims[0], deleteDims.size()));
+        lp_rm1.set("target_dims", DictValue::arrayInt(&targetDims[0], targetDims.size()));
+        int id_rm1 = net.addLayer(lp_rm1.name, lp_rm1.type, lp_rm1);
+        net.connect(id_pow, 0, id_rm1, 0);
+
+        Mat add_const(1, 1, CV_32F);
+        add_const.at<float>(0) = 1e-5;
+        LayerParams lp_add_const;
+        lp_add_const.type = "Const";
+        lp_add_const.name = "const2";
+        lp_add_const.blobs.push_back(add_const);
+        int id_add_const = net.addLayer(lp_add_const.name, lp_add_const.type, lp_add_const);
+        LayerParams lp_add;
+        lp_add.type = "NaryEltwise";
+        lp_add.name = "add1";
+        lp_add.set("operation", "add");
+        int id_add = net.addLayer(lp_add.name, lp_add.type, lp_add);
+        net.connect(id_rm1, 0, id_add, 0);
+        net.connect(id_add_const, 0, id_add, 1);
+
+        LayerParams lp_sqrt;
+        lp_sqrt.type = "Sqrt";
+        lp_sqrt.name = "sqrt1";
+        int id_sqrt = net.addLayer(lp_sqrt.name, lp_sqrt.type, lp_sqrt);
+        net.connect(id_add, 0, id_sqrt, 0);
+
+        LayerParams lp_div;
+        lp_div.type = "NaryEltwise";
+        lp_div.name = "div1";
+        lp_div.set("operation", "div");
+        int id_div = net.addLayer(lp_div.name, lp_div.type, lp_div);
+        net.connect(id_sub, 0, id_div, 0);
+        net.connect(id_sqrt, 0, id_div, 1);
+
+        LayerParams lp_mul;
+        lp_mul.type = "NaryEltwise";
+        lp_mul.name = "mul1";
+        lp_mul.set("operation", "mul");
+        int id_mul = net.addLayer(lp_mul.name, lp_mul.type, lp_mul);
+        net.connect(id_div, 0, id_mul, 0);
+        net.connect(0, 1, id_mul, 1);
+
+        LayerParams lp_add1;
+        lp_add1.type = "NaryEltwise";
+        lp_add1.name = "add2";
+        lp_add1.set("operation", "add");
+        int id_add1 = net.addLayer(lp_add1.name, lp_add1.type, lp_add1);
+        net.connect(id_mul, 0, id_add1, 0);
+        net.connect(0, 2, id_add1, 1);
+
+        // warmup
+        {
+            std::vector<String> inpNames(3);
+            inpNames[0] = "x";
+            inpNames[1] = "scale";
+            inpNames[2] = "b";
+            net.setInputsNames(inpNames);
+            net.setInput(x, inpNames[0]);
+            net.setInput(scale, inpNames[1]);
+            net.setInput(b, inpNames[2]);
+
+            net.setPreferableBackend(backendId);
+            net.setPreferableTarget(targetId);
+            Mat out = net.forward();
+        }
+
+        TEST_CYCLE()
+        {
+            Mat res = net.forward();
+        }
+
+        SANITY_CHECK_NOTHING();
+    }
+
+    int N = 1;
+    int H = 50;
+    int W = 768;
+};
+
+PERF_TEST_P_(Layer_LayerNormExpanded, DISABLED_LayerNormExpanded)
+{
+    test_layer({N, H ,W});
+}
+
 INSTANTIATE_TEST_CASE_P(/**/, Layer_Slice, dnnBackendsAndTargets(false, false));
 INSTANTIATE_TEST_CASE_P(/**/, Layer_NaryEltwise, testing::Values(std::make_tuple(DNN_BACKEND_OPENCV, DNN_TARGET_CPU)));
 #ifdef HAVE_CUDA
@@ -424,5 +630,7 @@ INSTANTIATE_TEST_CASE_P(CUDA, Layer_NaryEltwise, testing::Values(std::make_tuple
 #endif
 INSTANTIATE_TEST_CASE_P(/**/, Layer_Scatter, testing::Values(std::make_tuple(DNN_BACKEND_OPENCV, DNN_TARGET_CPU)));
 INSTANTIATE_TEST_CASE_P(/**/, Layer_ScatterND, testing::Values(std::make_tuple(DNN_BACKEND_OPENCV, DNN_TARGET_CPU)));
+INSTANTIATE_TEST_CASE_P(/**/, Layer_LayerNorm, testing::Values(std::make_tuple(DNN_BACKEND_OPENCV, DNN_TARGET_CPU)));
+INSTANTIATE_TEST_CASE_P(/**/, Layer_LayerNormExpanded, testing::Values(std::make_tuple(DNN_BACKEND_OPENCV, DNN_TARGET_CPU)));
 
 } // namespace

modules/dnn/src/cuda4dnn/primitives/matmul.hpp

@@ -38,8 +38,8 @@ namespace cv { namespace dnn { namespace cuda4dnn {
             const std::vector<cv::Ptr<BackendWrapper>>& outputs,
             csl::Workspace& workspace) override
         {
-            CV_Assert((inputs.size() == 2 && constTensor.empty() ||
-                       inputs.size() == 1 && !constTensor.empty()) && outputs.size() == 1);
+            CV_Assert(((inputs.size() == 2 && constTensor.empty()) ||
+                       (inputs.size() == 1 && !constTensor.empty())) && outputs.size() == 1);
 
             auto input1_wrapper = inputs[0].dynamicCast<wrapper_type>();
             auto input1 = input1_wrapper->getView();

modules/dnn/src/init.cpp

@@ -154,6 +154,7 @@ void initializeLayerFactory()
     CV_DNN_REGISTER_LAYER_CLASS(Arg,            ArgLayer);
     CV_DNN_REGISTER_LAYER_CLASS(Reciprocal,     ReciprocalLayer);
     CV_DNN_REGISTER_LAYER_CLASS(Gather,         GatherLayer);
+    CV_DNN_REGISTER_LAYER_CLASS(LayerNormalization, LayerNormLayer);
 
     CV_DNN_REGISTER_LAYER_CLASS(Crop,           CropLayer);
     CV_DNN_REGISTER_LAYER_CLASS(Eltwise,        EltwiseLayer);

modules/dnn/src/layers/fast_convolution/depthwise_convolution.cpp

@@ -29,7 +29,7 @@ static void depthWiseBlockConv2D(const float* wptr,
     const float w00_ = wptr[0], w01_ = wptr[1], w02_ = wptr[2],
             w10 = wptr[3], w11 = wptr[4], w12 = wptr[5],
             w20_ = wptr[6], w21_ = wptr[7], w22_ = wptr[8];
-    int outW1 = min(outW, (width - dilation_w*(kernel_w - 1) + pad_l)/stride_w);
+    const int outW1 = min(outW, (width - dilation_w*(kernel_w - 1) + pad_l)/stride_w);
     float relu_coeff = relu ? relu[out_d] : 1.f, bias = biasptr[out_d];
 
     for (int out_i = 0; out_i < outH; out_i++)
@@ -67,35 +67,37 @@ static void depthWiseBlockConv2D(const float* wptr,
 
 #if CV_SIMD128
         const int VEC_NLANES = 4;
-
-        if (fusedAdd)
-            outW1 = max(out_j, outW1 - outW1%VEC_NLANES);
-
-        v_float32x4 vw00 = v_setall_f32(w00);
-        v_float32x4 vw01 = v_setall_f32(w01);
-        v_float32x4 vw02 = v_setall_f32(w02);
-        v_float32x4 vw10 = v_setall_f32(w10);
-        v_float32x4 vw11 = v_setall_f32(w11);
-        v_float32x4 vw12 = v_setall_f32(w12);
-        v_float32x4 vw20 = v_setall_f32(w20);
-        v_float32x4 vw21 = v_setall_f32(w21);
-        v_float32x4 vw22 = v_setall_f32(w22);
-        v_float32x4 z = v_setzero_f32();
-        v_float32x4 vbias = v_setall_f32(bias);
-        v_float32x4 vrc = v_setall_f32(relu_coeff);
-
-        if (stride_w == 1 || (stride_w == 2 && dilation_w == 1))
+        if ((stride_w == 1 || (stride_w == 2 && dilation_w == 1)) && (outW1 - out_j) >= VEC_NLANES)
         {
-            if( stride_w == 1 )
+            v_float32x4 vw00 = v_setall_f32(w00);
+            v_float32x4 vw01 = v_setall_f32(w01);
+            v_float32x4 vw02 = v_setall_f32(w02);
+            v_float32x4 vw10 = v_setall_f32(w10);
+            v_float32x4 vw11 = v_setall_f32(w11);
+            v_float32x4 vw12 = v_setall_f32(w12);
+            v_float32x4 vw20 = v_setall_f32(w20);
+            v_float32x4 vw21 = v_setall_f32(w21);
+            v_float32x4 vw22 = v_setall_f32(w22);
+            v_float32x4 z = v_setzero_f32();
+            v_float32x4 vbias = v_setall_f32(bias);
+            v_float32x4 vrc = v_setall_f32(relu_coeff);
+
+            if (stride_w == 1)
             {
-                for( ; out_j < outW1; out_j += VEC_NLANES )
+                for (; out_j < outW1; out_j += VEC_NLANES)
                 {
-                    if (out_j + VEC_NLANES > outW1)
+                    // Tail processing.
+                    if (out_j > outW1 - VEC_NLANES)
                     {
-                        if (out_j <= pad_l || outW1 - VEC_NLANES < 0)
+                        // If fusedAdd is true, what is stored in outptr is not a meaningless value,
+                        // but the number being added. And we should avoid use tail processing in this case.
+                        // Because the tail process will make some elements compute twice,
+                        // which will lead to result errors.
+                        if (fusedAdd)
                             break;
                         out_j = outW1 - VEC_NLANES;
                     }
+
                     int in_j = out_j * stride_w - pad_l;
                     v_float32x4 v00 = v_load(imgptr0 + in_j),
                             v01 = v_load(imgptr0 + in_j + dilation_w),
@@ -119,11 +121,12 @@ static void depthWiseBlockConv2D(const float* wptr,
             }
             else // (stride_w == 2 && dilation_w == 1)
             {
-                for( ; out_j < outW1; out_j += VEC_NLANES )
+                for (; out_j < outW1; out_j += VEC_NLANES)
                 {
-                    if (out_j + VEC_NLANES > outW1 && out_j > pad_l)
+                    // Tail processing.
+                    if (out_j > outW1 - VEC_NLANES)
                     {
-                        if (outW1 - VEC_NLANES < 0)
+                        if (fusedAdd)
                             break;
                         out_j = outW1 - VEC_NLANES;
                     }
@@ -204,7 +207,7 @@ static void depthWiseBlockConv1D(const float* wptr,
                                  int out_d, int outW, bool fusedAdd)
 {
     const float w00_ = wptr[0], w01_ = wptr[1], w02_ = wptr[2];
-    int outW1 = min(outW, (width - dilation_w * (kernel_w - 1) + pad_l)/stride_w);
+    const int outW1 = min(outW, (width - dilation_w * (kernel_w - 1) + pad_l)/stride_w);
     float relu_coeff = relu ? relu[out_d] : 1.f, bias = biasptr[out_d];
 
     int out_j = 0;
@@ -225,27 +228,27 @@ static void depthWiseBlockConv1D(const float* wptr,
 
 #if CV_SIMD128
     const int VEC_NLANES = 4;
-    if (fusedAdd)
-        outW1 = max(out_j, outW1 - outW1%VEC_NLANES);
-    v_float32x4 vw00 = v_setall_f32(w00);
-    v_float32x4 vw01 = v_setall_f32(w01);
-    v_float32x4 vw02 = v_setall_f32(w02);
-    v_float32x4 z = v_setzero_f32();
-    v_float32x4 vbias = v_setall_f32(bias);
-    v_float32x4 vrc = v_setall_f32(relu_coeff);
-
-    if (stride_w == 1 || (stride_w == 2 && dilation_w == 1))
+    if ((stride_w == 1 || (stride_w == 2 && dilation_w == 1)) && (outW1 - out_j) >= VEC_NLANES)
     {
+        v_float32x4 vw00 = v_setall_f32(w00);
+        v_float32x4 vw01 = v_setall_f32(w01);
+        v_float32x4 vw02 = v_setall_f32(w02);
+        v_float32x4 z = v_setzero_f32();
+        v_float32x4 vbias = v_setall_f32(bias);
+        v_float32x4 vrc = v_setall_f32(relu_coeff);
+
         if( stride_w == 1 )
         {
             for( ; out_j < outW1; out_j += VEC_NLANES )
             {
+                // Tail processing.
                 if (out_j + VEC_NLANES > outW1)
                 {
-                    if (out_j <= pad_l || outW1 - VEC_NLANES < 0)
+                    if (fusedAdd)
                         break;
                     out_j = outW1 - VEC_NLANES;
                 }
+
                 int in_j = out_j * stride_w - pad_l;
                 v_float32x4 v00 = v_load(imgptr0 + in_j),
                         v01 = v_load(imgptr0 + in_j + dilation_w),
@@ -263,12 +266,14 @@ static void depthWiseBlockConv1D(const float* wptr,
         {
             for( ; out_j < outW1; out_j += VEC_NLANES )
             {
+                // Tail processing.
                 if (out_j + VEC_NLANES > outW1)
                 {
-                    if (out_j <= pad_l || outW1 - VEC_NLANES < 0)
+                    if (fusedAdd)
                         break;
                     out_j = outW1 - VEC_NLANES;
                 }
+
                 int in_j = out_j * stride_w - pad_l;
 
                 v_float32x4 v00, v01, v02, unused;

modules/dnn/src/layers/layer_norm.cpp

@@ -0,0 +1,176 @@
+// 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 "layers_common.hpp"
+
+namespace cv { namespace dnn {
+
+class LayerNormLayerImpl CV_FINAL : public LayerNormLayer
+{
+public:
+    LayerNormLayerImpl(const LayerParams& params)
+    {
+        setParamsFrom(params);
+
+        // standard attr
+        axis = params.get<int>("axis", 0);
+        epsilon = params.get<float>("epsilon", 1e-5);
+
+        // opencv attr
+        hasBias = params.get<bool>("hasBias", false);
+    }
+
+    virtual bool supportBackend(int backendId) CV_OVERRIDE
+    {
+        return backendId == DNN_BACKEND_OPENCV;
+    }
+
+    virtual bool getMemoryShapes(const std::vector<MatShape> &inputs,
+                                 const int requiredOutputs,
+                                 std::vector<MatShape> &outputs,
+                                 std::vector<MatShape> &internals) const CV_OVERRIDE
+    {
+        // check shapes of weight and bias if existed
+        // inputs >= 2 (X and Weight are requested, bias is optional)
+        CV_Check(inputs.size(), inputs.size() >= 2 && inputs.size() <= 3, "LayerNorm: require two (x, weight) or three (x, weight, bias) inputs");
+
+        auto x_shape = inputs[0];
+        int x_ndims = static_cast<int>(x_shape.size());
+
+        auto w_shape = inputs[1];
+        // if axis == last_dim, scale and b are both 1d tensor (represented as 2d mat nx1)
+        int w_ndims = static_cast<int>(w_shape.size());
+        w_ndims = (axis == x_ndims - 1 && w_ndims == 2) ? w_ndims - 1 : w_ndims;
+        CV_CheckEQ(x_ndims - axis, w_ndims, "LayerNorm: shape of weight does not match with given axis and shape of input");
+        for (int i = 0; i < w_ndims; ++i)
+            CV_CheckEQ(x_shape[axis+i], w_shape[i], "LayerNorm: weight dimensions does not match with input dimensions");
+        if (hasBias)
+        {
+            CV_CheckEQ(inputs.size(), (size_t)3, "");
+            auto b_shape = inputs[2];
+            CV_CheckEQ(w_shape.size(), b_shape.size(), "LayerNorm: shape of weight does not match with shape of bias");
+            for (size_t i = 0; i < w_shape.size(); ++i)
+                CV_CheckEQ(w_shape[i], b_shape[i], "LayerNorm: bias dimensions does not match with weight dimensions");
+        }
+
+        // only one output is needed; Mean & InvStdDev are not needed
+        // in inference and should beomitted in onnx importer
+        outputs.assign(1, inputs[0]);
+        return false;
+    }
+
+    template<bool hasBias>
+    class LayerNormInvoker : public ParallelLoopBody
+    {
+    public:
+        const Mat& src;
+        const float* scaleData;
+        const float* biasData;
+        Mat& dst;
+
+        float epsilon;
+
+        int total;
+        int normSize;
+        float invNormSize;
+
+        LayerNormInvoker(const Mat& src_, const Mat& scale, const Mat* b, Mat& dst_, int axis, float epsilon_)
+            : src(src_), scaleData(scale.ptr<float>()), biasData(nullptr), dst(dst_), epsilon(epsilon_)
+        {
+            if (hasBias)
+            {
+                CV_Assert(b != nullptr);
+                CV_Assert(b->isContinuous());
+                biasData = (const float*)b->ptr<float>();
+            }
+
+            auto dstShape = shape(dst);
+            total = std::accumulate(dstShape.begin(), dstShape.begin() + axis, 1, std::multiplies<int>());
+            normSize = std::accumulate(dstShape.begin() + axis, dstShape.end(), 1, std::multiplies<int>());
+            invNormSize = 1.0f / normSize;
+        }
+
+        static void run(const Mat& src, const Mat& scale, const Mat* b, Mat& dst, int axis, float epsilon)
+        {
+            CV_Assert(src.isContinuous());
+            CV_Assert(dst.isContinuous());
+            CV_CheckTypeEQ(src.type(), CV_32F, "DNN/LayerNorm: only support float32");
+            CV_CheckTypeEQ(src.type(), dst.type(), "");
+            CV_Assert(scale.isContinuous());
+
+            CV_CheckGE(epsilon, 0.0f, "");
+
+            LayerNormInvoker p(src, scale, b, dst, axis, epsilon);
+
+            double nstripes = ((size_t)p.total * p.normSize) * (1 / 1024.0);
+            // double nstripes = ((size_t)p.total) * (1 / 1024.0);
+            parallel_for_(Range(0, p.total), p, nstripes);
+        }
+
+        void operator()(const Range& r) const CV_OVERRIDE
+        {
+            int stripeStart = r.start;
+            int stripeEnd = r.end;
+
+            const float* srcData = src.ptr<float>();
+            float* dstData = dst.ptr<float>();
+
+            for (int ofs = stripeStart; ofs < stripeEnd; ++ofs)
+            {
+                const float* first = srcData + ofs * normSize;
+                float* dstFirst = dstData + ofs * normSize;
+
+                float mean = 0;
+                float meanSquare = 0;
+                for (int h = 0; h < normSize; ++h)
+                {
+                    float v = first[h];
+                    mean += v;
+                    meanSquare += v * v;
+                }
+                mean *= invNormSize;
+                meanSquare = std::sqrt(std::max(0.f, meanSquare * invNormSize - mean * mean) + epsilon);
+                float invMeanSquare = 1.0f / meanSquare;
+                for (int h = 0; h < normSize; ++h)
+                {
+                    float v = (first[h] - mean) * invMeanSquare * scaleData[h];
+                    if (hasBias) {
+                        v = v + biasData[h];
+                    }
+                    dstFirst[h] = v;
+                }
+            }
+        }
+    };
+
+    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());
+
+        if (inputs_arr.depth() == CV_16S)
+        {
+            forward_fallback(inputs_arr, outputs_arr, internals_arr);
+            return;
+        }
+
+        std::vector<Mat> inputs, outputs;
+        inputs_arr.getMatVector(inputs);
+        outputs_arr.getMatVector(outputs);
+
+        if (hasBias) {
+            LayerNormInvoker<true>::run(inputs[0], inputs[1], &inputs[2], outputs[0], axis, epsilon);
+        } else {
+            LayerNormInvoker<false>::run(inputs[0], inputs[1], nullptr, outputs[0], axis, epsilon);
+        }
+    }
+};
+
+Ptr<LayerNormLayer> LayerNormLayer::create(const LayerParams& params)
+{
+    return makePtr<LayerNormLayerImpl>(params);
+}
+
+}} // cv::dnn

modules/dnn/src/onnx/onnx_graph_simplifier.cpp

@@ -75,6 +75,28 @@ public:
         return makePtr<ONNXNodeWrapper>(node);
     }
 
+    int getInputInitializerId(int node_id, int node_input_id)
+    {
+        auto node = getNode(node_id);
+        std::string node_input_name = node->getInputName(node_input_id);
+        for (int i = 0; i < numInitializers; ++i)
+            if (net.initializer(i).name() == node_input_name)
+                return i;
+        CV_Error(Error::StsParseError, "Initializer with name " + node_input_name + " not found");
+    }
+
+    Mat getMatFromInitializer(int idx)
+    {
+        const opencv_onnx::TensorProto& tensor_proto = net.initializer(idx);
+        return getMatFromTensor(tensor_proto);
+    }
+
+    std::string getNameOfInitializer(int idx) const
+    {
+        const opencv_onnx::TensorProto& tensor_proto = net.initializer(idx);
+        return tensor_proto.name();
+    }
+
     virtual int getNumNodes() const CV_OVERRIDE
     {
         return numInputs + numInitializers + net.node_size();
@@ -110,6 +132,142 @@ private:
     opencv_onnx::GraphProto& net;
 };
 
+class LayerNormSubGraph : public Subgraph
+{
+public:
+    LayerNormSubGraph() : axis(-1), epsilon(1e-5)
+    {
+        //   -> ReduceMean ->     -> Pow(2) -> ReduceMean -> Add(epsilon) -> Sqrt ->
+        // x                  Sub                                                    Div -> Mul(scale) -> Add(bias)
+        //   --------------->     ------------------------------------------------->
+        // NOTE: Pow(2), Add(epsilon), Mul(scale), add(bias) can have constants as op_type Constant or Initializer
+        int input = addNodeToMatch("");
+        int mean = addNodeToMatch("ReduceMean", input);
+
+        int sub = addNodeToMatch("Sub", input, mean);
+
+        int pow = addNodeToMatch("Pow", sub, addNodeToMatch(""));
+        int mean1 = addNodeToMatch("ReduceMean", pow);
+        int add = addNodeToMatch("Add", mean1, addNodeToMatch(""));
+        int sqrt = addNodeToMatch("Sqrt", add);
+
+        int div = addNodeToMatch("Div", sub, sqrt);
+        int mul = addNodeToMatch("Mul", div, addNodeToMatch(""));
+        addNodeToMatch("Add", mul, addNodeToMatch(""));
+
+        setFusedNode("LayerNormalization", input);
+    }
+
+    static bool isWithInitializer(const std::vector<int>& matchedNodesIds)
+    {
+        // if node.getType() is Constant, Constant nodes are placed between other nodes
+        if (matchedNodesIds[2] - matchedNodesIds[1] != 1)
+            return false;
+        // if Initializer, there is no nodes for constant between other nodes
+        return true;
+    }
+
+    static float extractConstant(const Ptr<ImportGraphWrapper>& net, int node_id, int input_id, bool withInitializer)
+    {
+        if (withInitializer)
+        {
+            auto onnx_net = net.dynamicCast<ONNXGraphWrapper>();
+            int initializer_id = onnx_net->getInputInitializerId(node_id, input_id);
+            Mat const_mat = onnx_net->getMatFromInitializer(initializer_id);
+            return *const_mat.ptr<float>();
+        } else {
+            const Ptr<ImportNodeWrapper> node = net->getNode(node_id);
+            int constant_id = getInputNodeId(net, node, input_id);
+            Ptr<ImportNodeWrapper> constant_ptr = net->getNode(constant_id);
+            opencv_onnx::NodeProto* constant_node = constant_ptr.dynamicCast<ONNXNodeWrapper>()->node;
+            opencv_onnx::TensorProto constant_proto = constant_node->attribute(0).t();
+            Mat constant_mat = getMatFromTensor(constant_proto);
+            return *constant_mat.ptr<float>();
+        }
+    }
+
+    static float extractAxis(const Ptr<ImportGraphWrapper>& net, int node_id)
+    {
+        Ptr<ImportNodeWrapper> mean_ptr = net->getNode(node_id);
+        opencv_onnx::NodeProto* mean_node = mean_ptr.dynamicCast<ONNXNodeWrapper>()->node;
+        int axis_ = -1;
+        for (int i = 0; i < mean_node->attribute_size(); i++)
+        {
+            opencv_onnx::AttributeProto attr = mean_node->attribute(i);
+            if (attr.name() != "axes")
+                continue;
+            axis_ = static_cast<int>(attr.ints(0));
+        }
+        return axis_;
+    }
+
+    static std::string getInputName(const Ptr<ImportGraphWrapper>& net, int node_id, int input_id, bool withInitializer)
+    {
+        if (withInitializer)
+        {
+            auto onnx_net = net.dynamicCast<ONNXGraphWrapper>();
+            int initializer_id = onnx_net->getInputInitializerId(node_id, input_id);
+            return onnx_net->getNameOfInitializer(initializer_id);
+        } else {
+            const auto node = net->getNode(node_id);
+            return node->getInputName(input_id);
+        }
+    }
+
+    virtual bool match(const Ptr<ImportGraphWrapper>& net, int nodeId,
+                       std::vector<int>& matchedNodesIds,
+                       std::vector<int>& targetNodesIds) CV_OVERRIDE
+    {
+        if (Subgraph::match(net, nodeId, matchedNodesIds, targetNodesIds))
+        {
+            withInitializer = isWithInitializer(matchedNodesIds);
+
+            float pow_exp = extractConstant(net, matchedNodesIds[2], 1, withInitializer);
+            if (pow_exp - 2 > 1e-5) // not pow(2)
+                return false;
+
+            int axis_mean1 = extractAxis(net, matchedNodesIds[0]);
+            int axis_mean2 = extractAxis(net, matchedNodesIds[3]);
+            if (axis_mean1 != axis_mean2)
+                return false;
+            axis = axis_mean1;
+
+            epsilon = extractConstant(net, matchedNodesIds[4], 1, withInitializer);
+
+            weight_name = getInputName(net, matchedNodesIds[7], 1, withInitializer);
+            bias_name = getInputName(net, matchedNodesIds[8], 1, withInitializer);
+
+            return true;
+        }
+        return false;
+    }
+
+    virtual void finalize(const Ptr<ImportGraphWrapper>&,
+                          const Ptr<ImportNodeWrapper>& fusedNode,
+                          std::vector<Ptr<ImportNodeWrapper> >&) CV_OVERRIDE
+    {
+        opencv_onnx::NodeProto* node = fusedNode.dynamicCast<ONNXNodeWrapper>()->node;
+        // axis
+        opencv_onnx::AttributeProto* attr_axis = node->add_attribute();
+        attr_axis->set_name("axis");
+        attr_axis->set_i(axis);
+        // epsilon
+        opencv_onnx::AttributeProto* attr_epsilon = node->add_attribute();
+        attr_epsilon->set_name("epsilon");
+        attr_epsilon->set_f(epsilon);
+        // add input
+        node->add_input(weight_name);
+        node->add_input(bias_name);
+    }
+
+protected:
+    int axis;
+    float epsilon;
+    bool withInitializer;
+    std::string weight_name;
+    std::string bias_name;
+};
+
 class SoftMaxSubgraphBase : public Subgraph
 {
 public:
@@ -746,6 +904,7 @@ public:
 void simplifySubgraphs(opencv_onnx::GraphProto& net)
 {
     std::vector<Ptr<Subgraph> > subgraphs;
+    subgraphs.push_back(makePtr<LayerNormSubGraph>());
     subgraphs.push_back(makePtr<GatherCastSubgraph>());
     subgraphs.push_back(makePtr<MulCastSubgraph>());
     subgraphs.push_back(makePtr<UpsampleSubgraph>());

modules/dnn/src/onnx/onnx_importer.cpp

@@ -190,6 +190,7 @@ private:
     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);
+    void parseLayerNorm            (LayerParams& layerParams, const opencv_onnx::NodeProto& node_proto);
     void parseSimpleLayers         (LayerParams& layerParams, const opencv_onnx::NodeProto& node_proto);
 
     // Domain: com.microsoft
@@ -3285,6 +3286,56 @@ void ONNXImporter::parseTile(LayerParams& layerParams, const opencv_onnx::NodePr
     }
 }
 
+void ONNXImporter::parseLayerNorm(LayerParams& layerParams, const opencv_onnx::NodeProto& node_proto)
+{
+    // validate axis and convert it if negative
+    auto inputDims = static_cast<int>(outShapes[node_proto.input(0)].size());
+    int axis = layerParams.get<int>("axis", -1);
+    // axis: [-dims, dims)
+    CV_CheckGE(axis, -inputDims, "DNN/ONNXImporter: axis of LayerNormalization is out of range");
+    CV_CheckLT(axis,  inputDims, "DNN/ONNXImporter: axis of LayerNormalization is out of range");
+    axis = (axis + inputDims) % inputDims;
+    layerParams.set("axis", axis);
+
+    // check if bias existed
+    bool hasBias = false;
+    if (node_proto.input_size() > 2)
+        hasBias = true;
+    layerParams.set("hasBias", hasBias);
+
+    // constants as constant inputs
+    for (size_t i = 1; i < node_proto.input_size(); i++)
+    {
+        if (layer_id.find(node_proto.input(i)) == layer_id.end())
+        {
+            Mat blob = getBlob(node_proto, i);
+
+            LayerParams constParams;
+            constParams.name = node_proto.input(i);
+            constParams.type = "Const";
+            constParams.blobs.push_back(blob);
+
+            opencv_onnx::NodeProto proto;
+            proto.add_output(constParams.name);
+            addLayer(constParams, proto);
+        }
+    }
+
+    // Remove additional outputs (Mean, InvStdDev)
+    if (node_proto.output_size() > 1)
+    {
+        auto outputName = node_proto.output(0);
+        opencv_onnx::NodeProto node_proto_ = node_proto;
+        node_proto_.clear_output();
+        node_proto_.add_output(outputName);
+        addLayer(layerParams, node_proto_);
+    }
+    else
+    {
+        addLayer(layerParams, node_proto);
+    }
+}
+
 void ONNXImporter::parseSimpleLayers(LayerParams& layerParams, const opencv_onnx::NodeProto& node_proto)
 {
     bool is_all_input_const = true;
@@ -3987,6 +4038,7 @@ void ONNXImporter::buildDispatchMap_ONNX_AI(int opset_version)
     dispatch["SpaceToDepth"] = dispatch["DepthToSpace"] = &ONNXImporter::parseDepthToSpace;
     dispatch["ScatterElements"] = dispatch["Scatter"] = dispatch["ScatterND"] = &ONNXImporter::parseScatter;
     dispatch["Tile"] = &ONNXImporter::parseTile;
+    dispatch["LayerNormalization"] = &ONNXImporter::parseLayerNorm;
 
     dispatch["Equal"] = dispatch["Greater"] = dispatch["Less"] = dispatch["Pow"] = dispatch["Add"] =
             dispatch["Sub"] = dispatch["Mul"] = dispatch["Div"] = dispatch["GreaterOrEqual"] =

modules/dnn/test/test_int8_layers.cpp

@@ -974,6 +974,9 @@ TEST_P(Test_Int8_nets, opencv_face_detector)
 
 TEST_P(Test_Int8_nets, EfficientDet)
 {
+    if (cvtest::skipUnstableTests)
+        throw SkipTestException("Skip unstable test");  // detail: https://github.com/opencv/opencv/pull/23167
+
     applyTestTag(CV_TEST_TAG_DEBUG_VERYLONG);
     if (target == DNN_TARGET_OPENCL_FP16 && !ocl::Device::getDefault().isIntel())
         applyTestTag(CV_TEST_TAG_DNN_SKIP_OPENCL_FP16);

modules/dnn/test/test_onnx_importer.cpp

@@ -1731,6 +1731,11 @@ TEST_P(Test_ONNX_layers, DepthWiseAdd)
     testONNXModels("depthwiseconv_add");
 }
 
+TEST_P(Test_ONNX_layers, DepthStride2)
+{
+    testONNXModels("depthwise_stride2");
+}
+
 TEST_P(Test_ONNX_layers, SubFromConst)
 {
     testONNXModels("sub_from_const1");
@@ -2416,6 +2421,36 @@ TEST_P(Test_ONNX_layers, Tile)
     testONNXModels("tile", pb);
 }
 
+TEST_P(Test_ONNX_layers, LayerNorm)
+{
+    testONNXModels("test_layer_normalization_2d_axis0", pb, 0, 0, false, true, 3);
+    testONNXModels("test_layer_normalization_2d_axis1", pb, 0, 0, false, true, 3);
+    testONNXModels("test_layer_normalization_2d_axis_negative_1", pb, 0, 0, false, true, 3);
+    testONNXModels("test_layer_normalization_2d_axis_negative_2", pb, 0, 0, false, true, 3);
+    testONNXModels("test_layer_normalization_3d_axis0_epsilon", pb, 0, 0, false, true, 3);
+    testONNXModels("test_layer_normalization_3d_axis1_epsilon", pb, 0, 0, false, true, 3);
+    testONNXModels("test_layer_normalization_3d_axis2_epsilon", pb, 0, 0, false, true, 3);
+    testONNXModels("test_layer_normalization_3d_axis_negative_1_epsilon", pb, 0, 0, false, true, 3);
+    testONNXModels("test_layer_normalization_3d_axis_negative_2_epsilon", pb, 0, 0, false, true, 3);
+    testONNXModels("test_layer_normalization_3d_axis_negative_3_epsilon", pb, 0, 0, false, true, 3);
+    testONNXModels("test_layer_normalization_4d_axis0", pb, 0, 0, false, true, 3);
+    testONNXModels("test_layer_normalization_4d_axis1", pb, 0, 0, false, true, 3);
+    testONNXModels("test_layer_normalization_4d_axis2", pb, 0, 0, false, true, 3);
+    testONNXModels("test_layer_normalization_4d_axis3", pb, 0, 0, false, true, 3);
+    testONNXModels("test_layer_normalization_4d_axis_negative_1", pb, 0, 0, false, true, 3);
+    testONNXModels("test_layer_normalization_4d_axis_negative_2", pb, 0, 0, false, true, 3);
+    testONNXModels("test_layer_normalization_4d_axis_negative_3", pb, 0, 0, false, true, 3);
+    testONNXModels("test_layer_normalization_4d_axis_negative_4", pb, 0, 0, false, true, 3);
+    testONNXModels("test_layer_normalization_default_axis", pb, 0, 0, false, true, 3);
+}
+
+// for testing graph simplification
+TEST_P(Test_ONNX_layers, LayerNormExpanded)
+{
+    testONNXModels("layer_norm_expanded");
+    testONNXModels("layer_norm_expanded_with_initializers");
+}
+
 INSTANTIATE_TEST_CASE_P(/**/, Test_ONNX_nets, dnnBackendsAndTargets());
 
 }} // namespace

modules/features2d/src/sift.simd.hpp

@@ -960,11 +960,18 @@ if( dstMat.type() == CV_32F )
         __dst = v_min(v_max(v_cvt_f32(v_round(__dst * __nrm2)), __min), __max);
         v_store(dst + k, __dst);
     }
+#endif
+#if defined(__GNUC__) && __GNUC__ >= 9
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Waggressive-loop-optimizations"  // iteration XX invokes undefined behavior
 #endif
     for( ; k < len; k++ )
     {
         dst[k] = saturate_cast<uchar>(rawDst[k]*nrm2);
     }
+#if defined(__GNUC__) && __GNUC__ >= 9
+#pragma GCC diagnostic pop
+#endif
 }
 else // CV_8U
 {
@@ -984,9 +991,8 @@ else // CV_8U
 #endif
 
 #if defined(__GNUC__) && __GNUC__ >= 9
-// avoid warning "iteration 7 invokes undefined behavior" on Linux ARM64
 #pragma GCC diagnostic push
-#pragma GCC diagnostic ignored "-Waggressive-loop-optimizations"
+#pragma GCC diagnostic ignored "-Waggressive-loop-optimizations"  // iteration XX invokes undefined behavior
 #endif
     for( ; k < len; k++ )
     {

modules/features2d/test/test_descriptors_regression.cpp

@@ -82,7 +82,7 @@ TEST( Features2d_DescriptorExtractor, batch_ORB )
     for( i = 0; i < n; i++ )
     {
         string imgname = format("%s/img%d.png", path.c_str(), i+1);
-        Mat img = imread(imgname, 0);
+        Mat img = imread(imgname, IMREAD_GRAYSCALE);
         imgs.push_back(img);
     }
 
@@ -110,7 +110,7 @@ TEST( Features2d_DescriptorExtractor, batch_SIFT )
     for( i = 0; i < n; i++ )
     {
         string imgname = format("%s/img%d.png", path.c_str(), i+1);
-        Mat img = imread(imgname, 0);
+        Mat img = imread(imgname, IMREAD_GRAYSCALE);
         imgs.push_back(img);
     }
 

modules/imgcodecs/misc/java/test/ImgcodecsTest.java

@@ -45,7 +45,7 @@ public class ImgcodecsTest extends OpenCVTestCase {
     }
 
     public void testImreadStringInt() {
-        dst = Imgcodecs.imread(OpenCVTestRunner.LENA_PATH, 0);
+        dst = Imgcodecs.imread(OpenCVTestRunner.LENA_PATH, Imgcodecs.IMREAD_GRAYSCALE);
         assertFalse(dst.empty());
         assertEquals(1, dst.channels());
         assertTrue(512 == dst.cols());

modules/imgproc/src/distransform.cpp

@@ -740,10 +740,8 @@ void cv::distanceTransform( InputArray _src, OutputArray _dst, OutputArray _labe
     if( maskSize != cv::DIST_MASK_3 && maskSize != cv::DIST_MASK_5 && maskSize != cv::DIST_MASK_PRECISE )
         CV_Error( CV_StsBadSize, "Mask size should be 3 or 5 or 0 (precise)" );
 
-    if( distType == cv::DIST_C || distType == cv::DIST_L1 )
-        maskSize = !need_labels ? cv::DIST_MASK_3 : cv::DIST_MASK_5;
-    else if( distType == cv::DIST_L2 && need_labels )
-        maskSize = cv::DIST_MASK_5;
+    if ((distType == cv::DIST_C || distType == cv::DIST_L1) && !need_labels)
+        maskSize = cv::DIST_MASK_3;
 
     if( maskSize == cv::DIST_MASK_PRECISE )
     {

modules/imgproc/src/drawing.cpp

@@ -63,7 +63,7 @@ CollectPolyEdges( Mat& img, const Point2l* v, int npts,
                   int shift, Point offset=Point() );
 
 static void
-FillEdgeCollection( Mat& img, std::vector<PolyEdge>& edges, const void* color );
+FillEdgeCollection( Mat& img, std::vector<PolyEdge>& edges, const void* color, int line_type);
 
 static void
 PolyLine( Mat& img, const Point2l* v, int npts, bool closed,
@@ -1049,7 +1049,7 @@ EllipseEx( Mat& img, Point2l center, Size2l axes,
         v.push_back(center);
         std::vector<PolyEdge> edges;
         CollectPolyEdges( img,  &v[0], (int)v.size(), edges, color, line_type, XY_SHIFT );
-        FillEdgeCollection( img, edges, color );
+        FillEdgeCollection( img, edges, color, line_type );
     }
 }
 
@@ -1277,37 +1277,60 @@ CollectPolyEdges( Mat& img, const Point2l* v, int count, std::vector<PolyEdge>&
         pt1.x = (pt1.x + offset.x) << (XY_SHIFT - shift);
         pt1.y = (pt1.y + delta) >> shift;
 
-        if( line_type < cv::LINE_AA )
+        Point2l pt0c(pt0), pt1c(pt1);
+
+        if (line_type < cv::LINE_AA)
         {
             t0.y = pt0.y; t1.y = pt1.y;
             t0.x = (pt0.x + (XY_ONE >> 1)) >> XY_SHIFT;
             t1.x = (pt1.x + (XY_ONE >> 1)) >> XY_SHIFT;
-            Line( img, t0, t1, color, line_type );
+            Line(img, t0, t1, color, line_type);
+
+            // use clipped endpoints to create a more accurate PolyEdge
+            if ((unsigned)t0.x >= (unsigned)(img.cols) ||
+                (unsigned)t1.x >= (unsigned)(img.cols) ||
+                (unsigned)t0.y >= (unsigned)(img.rows) ||
+                (unsigned)t1.y >= (unsigned)(img.rows))
+            {
+                clipLine(img.size(), t0, t1);
+
+                if (t0.y != t1.y)
+                {
+                    pt0c.y = t0.y; pt1c.y = t1.y;
+                    pt0c.x = (int64)(t0.x) << XY_SHIFT;
+                    pt1c.x = (int64)(t1.x) << XY_SHIFT;
+                }
+            }
+            else
+            {
+                pt0c.x += XY_ONE >> 1;
+                pt1c.x += XY_ONE >> 1;
+            }
         }
         else
         {
             t0.x = pt0.x; t1.x = pt1.x;
             t0.y = pt0.y << XY_SHIFT;
             t1.y = pt1.y << XY_SHIFT;
-            LineAA( img, t0, t1, color );
+            LineAA(img, t0, t1, color);
         }
 
-        if( pt0.y == pt1.y )
+        if (pt0.y == pt1.y)
             continue;
 
-        if( pt0.y < pt1.y )
+        edge.dx = (pt1c.x - pt0c.x) / (pt1c.y - pt0c.y);
+        if (pt0.y < pt1.y)
         {
             edge.y0 = (int)(pt0.y);
             edge.y1 = (int)(pt1.y);
-            edge.x = pt0.x;
+            edge.x = pt0c.x + (pt0.y - pt0c.y) * edge.dx; // correct starting point for clipped lines
         }
         else
         {
             edge.y0 = (int)(pt1.y);
             edge.y1 = (int)(pt0.y);
-            edge.x = pt1.x;
+            edge.x = pt1c.x + (pt1.y - pt1c.y) * edge.dx; // correct starting point for clipped lines
         }
-        edge.dx = (pt1.x - pt0.x) / (pt1.y - pt0.y);
         edges.push_back(edge);
     }
 }
@@ -1324,7 +1347,7 @@ struct CmpEdges
 /**************** helper macros and functions for sequence/contour processing ***********/
 
 static void
-FillEdgeCollection( Mat& img, std::vector<PolyEdge>& edges, const void* color )
+FillEdgeCollection( Mat& img, std::vector<PolyEdge>& edges, const void* color, int line_type)
 {
     PolyEdge tmp;
     int i, y, total = (int)edges.size();
@@ -1333,6 +1356,12 @@ FillEdgeCollection( Mat& img, std::vector<PolyEdge>& edges, const void* color )
     int y_max = INT_MIN, y_min = INT_MAX;
     int64 x_max = 0xFFFFFFFFFFFFFFFF, x_min = 0x7FFFFFFFFFFFFFFF;
     int pix_size = (int)img.elemSize();
+    int delta;
+
+    if (line_type < LINE_AA)
+        delta = 0;
+    else
+        delta = XY_ONE - 1;
 
     if( total < 2 )
         return;
@@ -1411,12 +1440,12 @@ FillEdgeCollection( Mat& img, std::vector<PolyEdge>& edges, const void* color )
 
                     if (keep_prelast->x > prelast->x)
                     {
-                        x1 = (int)((prelast->x + XY_ONE - 1) >> XY_SHIFT);
+                        x1 = (int)((prelast->x + delta) >> XY_SHIFT);
                         x2 = (int)(keep_prelast->x >> XY_SHIFT);
                     }
                     else
                     {
-                        x1 = (int)((keep_prelast->x + XY_ONE - 1) >> XY_SHIFT);
+                        x1 = (int)((keep_prelast->x + delta) >> XY_SHIFT);
                         x2 = (int)(prelast->x >> XY_SHIFT);
                     }
 
@@ -2017,7 +2046,7 @@ void fillPoly( InputOutputArray _img, const Point** pts, const int* npts, int nc
         CollectPolyEdges(img, _pts.data(), npts[i], edges, buf, line_type, shift, offset);
     }
 
-    FillEdgeCollection(img, edges, buf);
+    FillEdgeCollection(img, edges, buf, line_type);
 }
 
 void polylines( InputOutputArray _img, const Point* const* pts, const int* npts, int ncontours, bool isClosed,
@@ -2353,7 +2382,7 @@ cvDrawContours( void* _img, CvSeq* contour,
     }
 
     if( thickness < 0 )
-        cv::FillEdgeCollection( img, edges, ext_buf );
+        cv::FillEdgeCollection( img, edges, ext_buf, line_type);
 
     if( h_next && contour0 )
         contour0->h_next = h_next;

modules/imgproc/test/test_connectedcomponents.cpp

@@ -81,7 +81,7 @@ void CV_ConnectedComponentsTest::run(int /* start_from */)
     int ccltype[] = { cv::CCL_DEFAULT, cv::CCL_WU, cv::CCL_GRANA, cv::CCL_BOLELLI, cv::CCL_SAUF, cv::CCL_BBDT, cv::CCL_SPAGHETTI };
 
     string exp_path = string(ts->get_data_path()) + "connectedcomponents/ccomp_exp.png";
-    Mat exp = imread(exp_path, 0);
+    Mat exp = imread(exp_path, IMREAD_GRAYSCALE);
     Mat orig = imread(string(ts->get_data_path()) + "connectedcomponents/concentric_circles.png", 0);
 
     if (orig.empty())

modules/imgproc/test/test_convhull.cpp

@@ -146,7 +146,7 @@ cvTsIsPointOnLineSegment(const cv::Point2f &x, const cv::Point2f &a, const cv::P
     double d2 = cvTsDist(x, b);
     double d3 = cvTsDist(a, b);
 
-    return (abs(d1 + d2 - d3) <= (1E-5));
+    return (abs(d1 + d2 - d3) <= (1E-4));
 }
 
 

modules/imgproc/test/test_drawing.cpp

@@ -857,4 +857,239 @@ TEST(Drawing, ttf_text)
 }
 #endif
 
+
+TEST(Drawing, fillpoly_fully)
+{
+    unsigned imageWidth = 256;
+    unsigned imageHeight = 256;
+    int type = CV_8UC1;
+    int shift = 0;
+    Point offset(0, 0);
+    cv::LineTypes lineType = LINE_4;
+
+    int imageSizeOffset = 15;
+
+    cv::Mat img(imageHeight, imageWidth, type);
+    img = 0;
+
+    std::vector<cv::Point> polygonPoints;
+    polygonPoints.push_back(cv::Point(100, -50));
+    polygonPoints.push_back(cv::Point(imageSizeOffset, imageHeight - imageSizeOffset));
+    polygonPoints.push_back(cv::Point(imageSizeOffset, imageSizeOffset));
+
+    // convert data
+    std::vector<const cv::Point*> polygonPointPointers(polygonPoints.size());
+    for (size_t i = 0; i < polygonPoints.size(); i++)
+    {
+        polygonPointPointers[i] = &polygonPoints[i];
+    }
+
+    const cv::Point** data = &polygonPointPointers.front();
+    int size = (int)polygonPoints.size();
+    const int* npts = &size;
+    int ncontours = 1;
+
+    // generate image
+    cv::fillPoly(img, data, npts, ncontours, 255, lineType, shift, offset);
+
+    // check for artifacts
+    {
+        cv::Mat binary = img < 128;
+        cv::Mat labelImage(binary.size(), CV_32S);
+        cv::Mat labelCentroids;
+        int labels = cv::connectedComponents(binary, labelImage, 4);
+        EXPECT_EQ(2, labels) << "artifacts occured";
+    }
+
+    // check if filling went over border
+    {
+        int xy_shift = 16, delta = offset.y + ((1 << shift) >> 1);
+        int xy_one = 1 << xy_shift;
+
+        Point pt0(polygonPoints[polygonPoints.size() - 1]), pt1;
+        for (size_t i = 0; i < polygonPoints.size(); i++, pt0 = pt1)
+        {
+            pt1 = polygonPoints[i];
+
+            // offset/shift treated like in fillPoly
+            Point t0(pt0), t1(pt1);
+
+            t0.x = (t0.x + offset.x) << (xy_shift - shift);
+            t0.y = (t0.y + delta) >> shift;
+
+            t1.x = (t1.x + offset.x) << (xy_shift - shift);
+            t1.y = (t1.y + delta) >> shift;
+
+            if (lineType < LINE_AA)
+            {
+                t0.x = (t0.x + (xy_one >> 1)) >> xy_shift;
+                t1.x = (t1.x + (xy_one >> 1)) >> xy_shift;
+
+                // LINE_4 to use the same type of line which is used in fillPoly
+                line(img, t0, t1, 0, 1, LINE_4, 0);
+            }
+            else
+            {
+                t0.x >>= (xy_shift);
+                t1.x >>= (xy_shift);
+                line(img, t0, t1, 0, 1, lineType, 0);
+            }
+
+        }
+        cv::Mat binary = img < 254;
+        cv::Mat labelImage(binary.size(), CV_32S);
+        int labels = cv::connectedComponents(binary, labelImage, 4);
+        EXPECT_EQ(2, labels) << "filling went over the border";
+    }
+}
+
+PARAM_TEST_CASE(FillPolyFully, unsigned, unsigned, int, int, Point, cv::LineTypes)
+{
+    unsigned imageWidth;
+    unsigned imageHeight;
+    int type;
+    int shift;
+    Point offset;
+    cv::LineTypes lineType;
+
+    virtual void SetUp()
+    {
+        imageWidth = GET_PARAM(0);
+        imageHeight = GET_PARAM(1);
+        type = GET_PARAM(2);
+        shift = GET_PARAM(3);
+        offset = GET_PARAM(4);
+        lineType = GET_PARAM(5);
+    }
+
+    void draw_polygon(cv::Mat& img, const std::vector<cv::Point>& polygonPoints)
+    {
+        // convert data
+        std::vector<const cv::Point*> polygonPointPointers(polygonPoints.size());
+        for (size_t i = 0; i < polygonPoints.size(); i++)
+        {
+            polygonPointPointers[i] = &polygonPoints[i];
+        }
+
+        const cv::Point** data = &polygonPointPointers.front();
+        int size = (int)polygonPoints.size();
+        const int* npts = &size;
+        int ncontours = 1;
+
+        // generate image
+        cv::fillPoly(img, data, npts, ncontours, 255, lineType, shift, offset);
+    }
+
+    void check_artifacts(cv::Mat& img)
+    {
+        // check for artifacts
+        cv::Mat binary = img < 128;
+        cv::Mat labelImage(binary.size(), CV_32S);
+        cv::Mat labelCentroids;
+        int labels = cv::connectedComponents(binary, labelImage, 4);
+        EXPECT_EQ(2, labels) << "artifacts occured";
+    }
+
+    void check_filling_over_border(cv::Mat& img, const std::vector<cv::Point>& polygonPoints)
+    {
+        int xy_shift = 16, delta = offset.y + ((1 << shift) >> 1);
+        int xy_one = 1 << xy_shift;
+
+        Point pt0(polygonPoints[polygonPoints.size() - 1]), pt1;
+        for (size_t i = 0; i < polygonPoints.size(); i++, pt0 = pt1)
+        {
+            pt1 = polygonPoints[i];
+
+            // offset/shift treated like in fillPoly
+            Point t0(pt0), t1(pt1);
+
+            t0.x = (t0.x + offset.x) << (xy_shift - shift);
+            t0.y = (t0.y + delta) >> shift;
+
+            t1.x = (t1.x + offset.x) << (xy_shift - shift);
+            t1.y = (t1.y + delta) >> shift;
+
+            if (lineType < LINE_AA)
+            {
+                t0.x = (t0.x + (xy_one >> 1)) >> xy_shift;
+                t1.x = (t1.x + (xy_one >> 1)) >> xy_shift;
+
+                // LINE_4 to use the same type of line which is used in fillPoly
+                line(img, t0, t1, 0, 1, LINE_4, 0);
+            }
+            else
+            {
+                t0.x >>= (xy_shift);
+                t1.x >>= (xy_shift);
+                line(img, t0, t1, 0, 1, lineType, 0);
+            }
+
+        }
+        cv::Mat binary = img < 254;
+        cv::Mat labelImage(binary.size(), CV_32S);
+        int labels = cv::connectedComponents(binary, labelImage, 4);
+        EXPECT_EQ(2, labels) << "filling went over the border";
+    }
+
+    void run_test(const std::vector<cv::Point>& polygonPoints)
+    {
+        cv::Mat img(imageHeight, imageWidth, type);
+        img = 0;
+
+        draw_polygon(img, polygonPoints);
+        check_artifacts(img);
+        check_filling_over_border(img, polygonPoints);
+    }
+};
+
+TEST_P(FillPolyFully, DISABLED_fillpoly_fully)
+{
+    int imageSizeOffset = 15;
+
+    // testing for polygon with straight edge at left/right side
+    int positions1[2] = { imageSizeOffset, (int)imageWidth - imageSizeOffset };
+    for (size_t i = 0; i < 2; i++)
+    {
+        for (int y = imageHeight + 50; y > -50; y -= 1)
+        {
+            // define polygon
+            std::vector<cv::Point> polygonPoints;
+            polygonPoints.push_back(cv::Point(100, imageHeight - y));
+            polygonPoints.push_back(cv::Point(positions1[i], positions1[1]));
+            polygonPoints.push_back(cv::Point(positions1[i], positions1[0]));
+
+            run_test(polygonPoints);
+        }
+    }
+
+    // testing for polygon with straight edge at top/bottom side
+    int positions2[2] = { imageSizeOffset, (int)imageHeight - imageSizeOffset };
+    for (size_t i = 0; i < 2; i++)
+    {
+        for (int x = imageWidth + 50; x > -50; x -= 1)
+        {
+            // define polygon
+            std::vector<cv::Point> polygonPoints;
+            polygonPoints.push_back(cv::Point(imageWidth - x, 100));
+            polygonPoints.push_back(cv::Point(positions2[1], positions2[i]));
+            polygonPoints.push_back(cv::Point(positions2[0], positions2[i]));
+
+            run_test(polygonPoints);
+        }
+    }
+}
+
+INSTANTIATE_TEST_CASE_P(
+    FillPolyTest, FillPolyFully,
+    testing::Combine(
+        testing::Values(256),
+        testing::Values(256),
+        testing::Values(CV_8UC1),
+        testing::Values(0, 1, 2),
+        testing::Values(cv::Point(0, 0), cv::Point(10, 10)),
+        testing::Values(LINE_4, LINE_8, LINE_AA)
+    )
+);
+
+
 }} // namespace

modules/imgproc/test/test_imgproc_umat.cpp

@@ -53,7 +53,7 @@ protected:
     void run(int)
     {
         string imgpath = string(ts->get_data_path()) + "shared/lena.png";
-        Mat img = imread(imgpath, 1), gray, smallimg, result;
+        Mat img = imread(imgpath, IMREAD_COLOR), gray, smallimg, result;
         UMat uimg = img.getUMat(ACCESS_READ), ugray, usmallimg, uresult;
 
         cvtColor(img, gray, COLOR_BGR2GRAY);

modules/imgproc/test/test_imgwarp.cpp

@@ -102,7 +102,6 @@ void CV_ImgWarpBaseTest::get_test_array_types_and_sizes( int test_case_idx,
     int cn = cvtest::randInt(rng) % 3 + 1;
     cvtest::ArrayTest::get_test_array_types_and_sizes( test_case_idx, sizes, types );
     depth = depth == 0 ? CV_8U : depth == 1 ? CV_16U : CV_32F;
-    cn += cn == 2;
 
     types[INPUT][0] = types[INPUT_OUTPUT][0] = types[REF_INPUT_OUTPUT][0] = CV_MAKETYPE(depth, cn);
     if( test_array[INPUT].size() > 1 )

modules/imgproc/test/test_imgwarp_strict.cpp

@@ -151,8 +151,6 @@ void CV_ImageWarpBaseTest::generate_test_data()
         depth = rng.uniform(0, CV_64F);
 
     int cn = rng.uniform(1, 4);
-    while (cn == 2)
-        cn = rng.uniform(1, 4);
 
     src.create(ssize, CV_MAKE_TYPE(depth, cn));
 
@@ -237,7 +235,7 @@ float CV_ImageWarpBaseTest::get_success_error_level(int _interpolation, int) con
     else if (_interpolation == INTER_LANCZOS4)
         return 1.0f;
     else if (_interpolation == INTER_NEAREST)
-        return 1.0f;
+        return 255.0f;  // FIXIT: check is not reliable for Black/White (0/255) images
     else if (_interpolation == INTER_AREA)
         return 2.0f;
     else
@@ -430,8 +428,6 @@ void CV_Resize_Test::generate_test_data()
         depth = rng.uniform(0, CV_64F);
 
     int cn = rng.uniform(1, 4);
-    while (cn == 2)
-        cn = rng.uniform(1, 4);
 
     src.create(ssize, CV_MAKE_TYPE(depth, cn));
 

modules/java/test/android_test/src/org/opencv/test/OpenCVTestCase.java

@@ -149,7 +149,7 @@ public class OpenCVTestCase extends TestCase {
         rgba128 = new Mat(matSize, matSize, CvType.CV_8UC4, Scalar.all(128));
 
         rgbLena = Imgcodecs.imread(OpenCVTestRunner.LENA_PATH);
-        grayChess = Imgcodecs.imread(OpenCVTestRunner.CHESS_PATH, 0);
+        grayChess = Imgcodecs.imread(OpenCVTestRunner.CHESS_PATH, Imgcodecs.IMREAD_GRAYSCALE);
 
         gray255_32f_3d = new Mat(new int[]{matSize, matSize, matSize}, CvType.CV_32F, new Scalar(255.0));
 

modules/java/test/pure_test/src/org/opencv/test/OpenCVTestCase.java

@@ -175,7 +175,7 @@ public class OpenCVTestCase extends TestCase {
         rgba128 = new Mat(matSize, matSize, CvType.CV_8UC4, Scalar.all(128));
 
         rgbLena = Imgcodecs.imread(OpenCVTestRunner.LENA_PATH);
-        grayChess = Imgcodecs.imread(OpenCVTestRunner.CHESS_PATH, 0);
+        grayChess = Imgcodecs.imread(OpenCVTestRunner.CHESS_PATH, Imgcodecs.IMREAD_GRAYSCALE);
 
         gray255_32f_3d = new Mat(new int[]{matSize, matSize, matSize}, CvType.CV_32F, new Scalar(255.0));
 

modules/objdetect/test/test_cascadeandhog.cpp

@@ -138,7 +138,7 @@ int CV_DetectorTest::prepareData( FileStorage& _fs )
                 String filename;
                 it >> filename;
                 imageFilenames.push_back(filename);
-                Mat img = imread( dataPath+filename, 1 );
+                Mat img = imread( dataPath+filename, IMREAD_COLOR );
                 images.push_back( img );
             }
         }

modules/photo/include/opencv2/photo.hpp

@@ -201,8 +201,8 @@ CV_EXPORTS_W void fastNlMeansDenoisingColored( InputArray src, OutputArray dst,
 
 /** @brief Modification of fastNlMeansDenoising function for images sequence where consecutive images have been
 captured in small period of time. For example video. This version of the function is for grayscale
-images or for manual manipulation with colorspaces. For more details see
-<http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.131.6394>
+images or for manual manipulation with colorspaces. See @cite Buades2005DenoisingIS for more details
+(open access [here](https://static.aminer.org/pdf/PDF/000/317/196/spatio_temporal_wiener_filtering_of_image_sequences_using_a_parametric.pdf)).
 
 @param srcImgs Input 8-bit 1-channel, 2-channel, 3-channel or
 4-channel images sequence. All images should have the same type and
@@ -228,8 +228,8 @@ CV_EXPORTS_W void fastNlMeansDenoisingMulti( InputArrayOfArrays srcImgs, OutputA
 
 /** @brief Modification of fastNlMeansDenoising function for images sequence where consecutive images have been
 captured in small period of time. For example video. This version of the function is for grayscale
-images or for manual manipulation with colorspaces. For more details see
-<http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.131.6394>
+images or for manual manipulation with colorspaces. See @cite Buades2005DenoisingIS for more details
+(open access [here](https://static.aminer.org/pdf/PDF/000/317/196/spatio_temporal_wiener_filtering_of_image_sequences_using_a_parametric.pdf)).
 
 @param srcImgs Input 8-bit or 16-bit (only with NORM_L1) 1-channel,
 2-channel, 3-channel or 4-channel images sequence. All images should

modules/photo/test/test_denoising.cpp

@@ -157,7 +157,7 @@ TEST(Photo_White, issue_2646)
 TEST(Photo_Denoising, speed)
 {
     string imgname = string(cvtest::TS::ptr()->get_data_path()) + "shared/5MP.png";
-    Mat src = imread(imgname, 0), dst;
+    Mat src = imread(imgname, IMREAD_GRAYSCALE), dst;
 
     double t = (double)getTickCount();
     fastNlMeansDenoising(src, dst, 5, 7, 21);

modules/stereo/test/test_stereomatching.cpp

@@ -456,8 +456,8 @@ void CV_StereoMatchingTest::run(int)
         string datasetFullDirName = dataPath + DATASETS_DIR + datasetName + "/";
         Mat leftImg = imread(datasetFullDirName + LEFT_IMG_NAME);
         Mat rightImg = imread(datasetFullDirName + RIGHT_IMG_NAME);
-        Mat trueLeftDisp = imread(datasetFullDirName + TRUE_LEFT_DISP_NAME, 0);
-        Mat trueRightDisp = imread(datasetFullDirName + TRUE_RIGHT_DISP_NAME, 0);
+        Mat trueLeftDisp = imread(datasetFullDirName + TRUE_LEFT_DISP_NAME, IMREAD_GRAYSCALE);
+        Mat trueRightDisp = imread(datasetFullDirName + TRUE_RIGHT_DISP_NAME, IMREAD_GRAYSCALE);
         Rect calcROI;
 
         if( leftImg.empty() || rightImg.empty() || trueLeftDisp.empty() )
@@ -835,9 +835,9 @@ TEST_P(Calib3d_StereoBM_BufferBM, memAllocsTest)
     const int SADWindowSize = get<1>(get<1>(GetParam()));
 
     String path = cvtest::TS::ptr()->get_data_path() + "cv/stereomatching/datasets/teddy/";
-    Mat leftImg = imread(path + "im2.png", 0);
+    Mat leftImg = imread(path + "im2.png", IMREAD_GRAYSCALE);
     ASSERT_FALSE(leftImg.empty());
-    Mat rightImg = imread(path + "im6.png", 0);
+    Mat rightImg = imread(path + "im6.png", IMREAD_GRAYSCALE);
     ASSERT_FALSE(rightImg.empty());
     Mat leftDisp;
     {
@@ -923,9 +923,9 @@ TEST(Calib3d_StereoSGBM, regression) { CV_StereoSGBMTest test; test.safe_run();
 TEST(Calib3d_StereoSGBM_HH4, regression)
 {
     String path = cvtest::TS::ptr()->get_data_path() + "cv/stereomatching/datasets/teddy/";
-    Mat leftImg = imread(path + "im2.png", 0);
+    Mat leftImg = imread(path + "im2.png", IMREAD_GRAYSCALE);
     ASSERT_FALSE(leftImg.empty());
-    Mat rightImg = imread(path + "im6.png", 0);
+    Mat rightImg = imread(path + "im6.png", IMREAD_GRAYSCALE);
     ASSERT_FALSE(rightImg.empty());
     Mat testData = imread(path + "disp2_hh4.png",-1);
     ASSERT_FALSE(testData.empty());

modules/ts/include/opencv2/ts/ts_perf.hpp

@@ -454,9 +454,6 @@ private:
     performance_metrics metrics;
     void validateMetrics();
 
-    static int64 _timeadjustment;
-    static int64 _calibrate();
-
     static void warmup_impl(cv::Mat m, WarmUpType wtype);
     static int getSizeInBytes(cv::InputArray a);
     static cv::Size getSize(cv::InputArray a);

modules/ts/src/cuda_test.cpp

@@ -522,13 +522,35 @@ namespace cvtest
 
         int validCount = 0;
 
-        for (size_t i = 0; i < gold.size(); ++i)
+        if (actual.size() == gold.size())
         {
-            const cv::KeyPoint& p1 = gold[i];
-            const cv::KeyPoint& p2 = actual[i];
+            for (size_t i = 0; i < gold.size(); ++i)
+            {
+                const cv::KeyPoint& p1 = gold[i];
+                const cv::KeyPoint& p2 = actual[i];
 
-            if (keyPointsEquals(p1, p2))
-                ++validCount;
+                if (keyPointsEquals(p1, p2))
+                    ++validCount;
+            }
+        }
+        else
+        {
+            std::vector<cv::KeyPoint>& shorter = gold;
+            std::vector<cv::KeyPoint>& longer = actual;
+            if (actual.size() < gold.size())
+            {
+                shorter = actual;
+                longer = gold;
+            }
+            for (size_t i = 0; i < shorter.size(); ++i)
+            {
+                const cv::KeyPoint& p1 = shorter[i];
+                const cv::KeyPoint& p2 = longer[i];
+                const cv::KeyPoint& p3 = longer[i+1];
+
+                if (keyPointsEquals(p1, p2) || keyPointsEquals(p1, p3))
+                    ++validCount;
+            }
         }
 
         return validCount;

modules/ts/src/ts.cpp

@@ -623,20 +623,27 @@ void TS::set_gtest_status()
 
 void TS::update_context( BaseTest* test, int test_case_idx, bool update_ts_context )
 {
+    CV_UNUSED(update_ts_context);
+
     if( current_test_info.test != test )
     {
         for( int i = 0; i <= CONSOLE_IDX; i++ )
             output_buf[i] = string();
-        rng = RNG(params.rng_seed);
-        current_test_info.rng_seed0 = current_test_info.rng_seed = rng.state;
+    }
+
+    if (test_case_idx >= 0)
+    {
+        current_test_info.rng_seed = param_seed + test_case_idx;
+        current_test_info.rng_seed0 = current_test_info.rng_seed;
+
+        rng = RNG(current_test_info.rng_seed);
+        cv::theRNG() = rng;
     }
 
     current_test_info.test = test;
     current_test_info.test_case_idx = test_case_idx;
     current_test_info.code = 0;
     cvSetErrStatus( CV_StsOk );
-    if( update_ts_context )
-        current_test_info.rng_seed = rng.state;
 }
 
 

modules/ts/src/ts_perf.cpp

@@ -26,7 +26,6 @@ using namespace perf;
 
 int64 TestBase::timeLimitDefault = 0;
 unsigned int TestBase::iterationsLimitDefault = UINT_MAX;
-int64 TestBase::_timeadjustment = 0;
 
 // Item [0] will be considered the default implementation.
 static std::vector<std::string> available_impls;
@@ -1159,7 +1158,6 @@ void TestBase::Init(const std::vector<std::string> & availableImpls,
 
     timeLimitDefault = param_time_limit == 0.0 ? 1 : (int64)(param_time_limit * cv::getTickFrequency());
     iterationsLimitDefault = param_force_samples == 0 ? UINT_MAX : param_force_samples;
-    _timeadjustment = _calibrate();
 }
 
 void TestBase::RecordRunParameters()
@@ -1193,66 +1191,6 @@ enum PERF_STRATEGY TestBase::getCurrentModulePerformanceStrategy()
     return strategyForce == PERF_STRATEGY_DEFAULT ? strategyModule : strategyForce;
 }
 
-
-int64 TestBase::_calibrate()
-{
-    CV_TRACE_FUNCTION();
-    if (iterationsLimitDefault <= 1)
-        return 0;
-
-    class _helper : public ::perf::TestBase
-    {
-    public:
-        _helper() { testStrategy = PERF_STRATEGY_BASE; }
-        performance_metrics& getMetrics() { return calcMetrics(); }
-        virtual void TestBody() {}
-        virtual void PerfTestBody()
-        {
-            //the whole system warmup
-            SetUp();
-            cv::Mat a(2048, 2048, CV_32S, cv::Scalar(1));
-            cv::Mat b(2048, 2048, CV_32S, cv::Scalar(2));
-            declare.time(30);
-            double s = 0;
-            declare.iterations(20);
-            minIters = nIters = 20;
-            for(; next() && startTimer(); stopTimer())
-                s+=a.dot(b);
-            declare.time(s);
-
-            //self calibration
-            SetUp();
-            declare.iterations(1000);
-            minIters = nIters = 1000;
-            for(int iters = 0; next() && startTimer(); iters++, stopTimer()) { /*std::cout << iters << nIters << std::endl;*/ }
-        }
-    };
-
-    // Initialize ThreadPool
-    class _dummyParallel : public ParallelLoopBody
-    {
-    public:
-       void operator()(const cv::Range& range) const
-       {
-           // nothing
-           CV_UNUSED(range);
-       }
-    };
-    parallel_for_(cv::Range(0, 1000), _dummyParallel());
-
-    _timeadjustment = 0;
-    _helper h;
-    h.PerfTestBody();
-    double compensation = h.getMetrics().min;
-    if (getCurrentModulePerformanceStrategy() == PERF_STRATEGY_SIMPLE)
-    {
-        CV_Assert(compensation < 0.01 * cv::getTickFrequency());
-        compensation = 0.0f; // simple strategy doesn't require any compensation
-    }
-    LOGD("Time compensation is %.0f", compensation);
-    return (int64)compensation;
-}
-
 #ifdef _MSC_VER
 # pragma warning(push)
 # pragma warning(disable:4355)  // 'this' : used in base member initializer list
@@ -1561,9 +1499,8 @@ void TestBase::stopTimer()
     if (lastTime == 0)
         ADD_FAILURE() << "  stopTimer() is called before startTimer()/next()";
     lastTime = time - lastTime;
+    CV_Assert(lastTime >= 0);  // TODO: CV_Check* for int64
     totalTime += lastTime;
-    lastTime -= _timeadjustment;
-    if (lastTime < 0) lastTime = 0;
     times.push_back(lastTime);
     lastTime = 0;
 

modules/video/test/test_estimaterigid.cpp

@@ -83,7 +83,7 @@ struct WrapAff2D
 
 bool CV_RigidTransform_Test::testNPoints(int from)
 {
-    cv::RNG rng = ts->get_rng();
+    cv::RNG rng = cv::theRNG();
 
     int progress = 0;
     int k, ntests = 10000;

modules/videoio/src/cap_msmf.cpp

@@ -2469,6 +2469,12 @@ const GUID CvVideoWriter_MSMF::FourCC2GUID(int fourcc)
 #endif
         case CV_FOURCC_MACRO('H', '2', '6', '4'):
                 return MFVideoFormat_H264; break;
+#if defined(NTDDI_WIN10)
+        case CV_FOURCC_MACRO('H', '2', '6', '5'):
+                return MFVideoFormat_H265;  break;
+        case CV_FOURCC_MACRO('H', 'E', 'V', 'C'):
+                return MFVideoFormat_HEVC;  break;
+#endif
         case CV_FOURCC_MACRO('M', '4', 'S', '2'):
                 return MFVideoFormat_M4S2; break;
         case CV_FOURCC_MACRO('M', 'J', 'P', 'G'):

modules/videoio/test/test_video_io.cpp

@@ -388,6 +388,7 @@ static Ext_Fourcc_PSNR synthetic_params[] = {
     {"wmv", "WMV3", 30.f, CAP_MSMF},
     {"wmv", "WVC1", 30.f, CAP_MSMF},
     {"mov", "H264", 30.f, CAP_MSMF},
+ // {"mov", "HEVC", 30.f, CAP_MSMF},  // excluded due to CI issue: https://github.com/opencv/opencv/pull/23172
 #endif
 
 #ifdef HAVE_AVFOUNDATION
@@ -995,6 +996,7 @@ static Ext_Fourcc_PSNR hw_codecs[] = {
 #ifdef _WIN32
         {"mp4", "MPEG", 29.f, CAP_MSMF},
         {"mp4", "H264", 29.f, CAP_MSMF},
+        {"mp4", "HEVC", 29.f, CAP_MSMF},
 #endif
 };
 

samples/cpp/3calibration.cpp

@@ -252,7 +252,7 @@ int main( int argc, char** argv )
         {
             int k1 = k == 0 ? 2 : k == 1 ? 0 : 1;
             printf("%s\n", imageList[i*3+k].c_str());
-            view = imread(imageList[i*3+k], 1);
+            view = imread(imageList[i*3+k], IMREAD_COLOR);
 
             if(!view.empty())
             {
@@ -340,7 +340,7 @@ int main( int argc, char** argv )
         {
             int k1 = k == 0 ? 2 : k == 1 ? 0 : 1;
             int k2 = k == 0 ? 1 : k == 1 ? 0 : 2;
-            view = imread(imageList[i*3+k], 1);
+            view = imread(imageList[i*3+k], IMREAD_COLOR);
 
             if(view.empty())
                 continue;

samples/cpp/calibration.cpp

@@ -497,7 +497,7 @@ int main( int argc, char** argv )
             view0.copyTo(view);
         }
         else if( i < (int)imageList.size() )
-            view = imread(imageList[i], 1);
+            view = imread(imageList[i], IMREAD_COLOR);
 
         if(view.empty())
         {
@@ -622,7 +622,7 @@ int main( int argc, char** argv )
 
         for( i = 0; i < (int)imageList.size(); i++ )
         {
-            view = imread(imageList[i], 1);
+            view = imread(imageList[i], IMREAD_COLOR);
             if(view.empty())
                 continue;
             remap(view, rview, map1, map2, INTER_LINEAR);

samples/cpp/facedetect.cpp

@@ -145,7 +145,7 @@ int main( int argc, const char** argv )
                         len--;
                     buf[len] = '\0';
                     cout << "file " << buf << endl;
-                    image = imread( buf, 1 );
+                    image = imread( buf, IMREAD_COLOR );
                     if( !image.empty() )
                     {
                         detectAndDraw( image, cascade, nestedCascade, scale, tryflip );

samples/cpp/pca.cpp

@@ -59,7 +59,7 @@ static void read_imgList(const string& filename, vector<Mat>& images) {
     }
     string line;
     while (getline(file, line)) {
-        images.push_back(imread(line, 0));
+        images.push_back(imread(line, IMREAD_GRAYSCALE));
     }
 }
 

samples/cpp/stereo_calib.cpp

@@ -82,7 +82,7 @@ StereoCalib(const vector<string>& imagelist, Size boardSize, float squareSize, b
         for( k = 0; k < 2; k++ )
         {
             const string& filename = imagelist[i*2+k];
-            Mat img = imread(filename, 0);
+            Mat img = imread(filename, IMREAD_GRAYSCALE);
             if(img.empty())
                 break;
             if( imageSize == Size() )
@@ -300,7 +300,7 @@ StereoCalib(const vector<string>& imagelist, Size boardSize, float squareSize, b
     {
         for( k = 0; k < 2; k++ )
         {
-            Mat img = imread(goodImageList[i*2+k], 0), rimg, cimg;
+            Mat img = imread(goodImageList[i*2+k], IMREAD_GRAYSCALE), rimg, cimg;
             remap(img, rimg, rmap[k][0], rmap[k][1], INTER_LINEAR);
             cvtColor(rimg, cimg, COLOR_GRAY2BGR);
             Mat canvasPart = !isVerticalStereo ? canvas(Rect(w*k, 0, w, h)) : canvas(Rect(0, h*k, w, h));

samples/cpp/tutorial_code/snippets/imgproc_HoughLinesCircles.cpp

@@ -8,7 +8,7 @@ using namespace std;
 int main(int argc, char** argv)
 {
     Mat img, gray;
-    if( argc != 2 || !(img=imread(argv[1], 1)).data)
+    if( argc != 2 || !(img=imread(argv[1], IMREAD_COLOR)).data)
         return -1;
     cvtColor(img, gray, COLOR_BGR2GRAY);
     // smooth it, otherwise a lot of false circles may be detected

samples/cpp/tutorial_code/snippets/imgproc_HoughLinesP.cpp

@@ -7,7 +7,7 @@ using namespace std;
 int main(int argc, char** argv)
 {
     Mat src, dst, color_dst;
-    if( argc != 2 || !(src=imread(argv[1], 0)).data)
+    if( argc != 2 || !(src=imread(argv[1], IMREAD_GRAYSCALE)).data)
         return -1;
 
     Canny( src, dst, 50, 200, 3 );

samples/cpp/tutorial_code/snippets/imgproc_calcHist.cpp

@@ -6,7 +6,7 @@ using namespace cv;
 int main( int argc, char** argv )
 {
     Mat src, hsv;
-    if( argc != 2 || !(src=imread(argv[1], 1)).data )
+    if( argc != 2 || !(src=imread(argv[1], IMREAD_COLOR)).data )
         return -1;
 
     cvtColor(src, hsv, COLOR_BGR2HSV);

samples/cpp/tutorial_code/snippets/imgproc_drawContours.cpp

@@ -9,7 +9,7 @@ int main( int argc, char** argv )
     Mat src;
     // the first command-line parameter must be a filename of the binary
     // (black-n-white) image
-    if( argc != 2 || !(src=imread(argv[1], 0)).data)
+    if( argc != 2 || !(src=imread(argv[1], IMREAD_GRAYSCALE)).data)
         return -1;
 
     Mat dst = Mat::zeros(src.rows, src.cols, CV_8UC3);

samples/cpp/watershed.cpp

@@ -54,7 +54,7 @@ int main( int argc, char** argv )
         return 0;
     }
     string filename = samples::findFile(parser.get<string>("@input"));
-    Mat img0 = imread(filename, 1), imgGray;
+    Mat img0 = imread(filename, IMREAD_COLOR), imgGray;
 
     if( img0.empty() )
     {

samples/python/calibrate.py

@@ -57,7 +57,7 @@ def main():
 
     def processImage(fn):
         print('processing %s... ' % fn)
-        img = cv.imread(fn, 0)
+        img = cv.imread(fn, cv.IMREAD_GRAYSCALE)
         if img is None:
             print("Failed to load", fn)
             return None

samples/python/mouse_and_match.py

@@ -69,7 +69,7 @@ class App():
             if ext == "png" or ext == "jpg" or ext == "bmp" or ext == "tiff" or ext == "pbm":
                 print(infile)
 
-                img = cv.imread(infile,1)
+                img = cv.imread(infile, cv.IMREAD_COLOR)
                 if img is None:
                     continue
                 self.sel = (0,0,0,0)