opencv_contrib · Issues· 722 open

EMD L1 interpreting every input matrix as 1-dimensional data

CUDA SeparableLinearFilter might fail because of too large grid dimensions

cudaarithm: cuda::multiply / sqr compute the 16U product in int and do not saturate (65535*65535 = 0); a 32S product with dtype=CV_32F is wrapped too

cudaarithm: cuda::abs does not saturate the most negative 8S/16S/32S value (abs(-128) = -128), and cuda::norm(NORM_INF) can return 0 or a negative value for such inputs

cudaarithm: bitwise_and/or/xor/not on a GpuMat ROI whose data pointer is not 4-byte (2-byte) aligned fail with a sticky "misaligned address" CUDA error

cudaimgproc: calcHist / histEven / equalizeHist on a GpuMat ROI with x offset > 4 count pixels outside the ROI (the offsetX handling from #3475 only works for offsets 0..4)

cudaimgproc: cuda::demosaicing (and cuda::cvtColor with Bayer codes) on a GpuMat ROI whose x offset is not a multiple of 4 (8U) / 2 (16U) pixels fails with a sticky "misaligned address" CUDA error

EdgeDrawing::ValidateCircles() reads an uninitialised ellipse point for odd perimeters

EdgeDrawing segmentation fault

EdgeDrawing::detectEllipses() modifies the caller's input image when Params::Sigma < 1.0

optflow: calcOpticalFlowSparseToDense segfaults when k exceeds the number of sparse matches

wechat qrcode UnicodeDecodeError in python3.6.9

CI: PR-5.x workflow fails for all PRs - references deleted geometry_module branch

Proposal: CUDA-accelerated implementation of TrackerCSRT

Incorrect documentation for WeChatQRCode module