Hacking OpenCV (Eng)
29 May 2018This post will show a subtle bug in OpenCV. In this post, version 3.4.1, which is the latest released version at the time of writing, is used.
Let’s look at the code above. Between line 23 and line 29, there are no codes that modify the value of GpuMat d2. Therefore, two identical images should be displayed on windows out1 and out2. However, different images are displayed on those windows as shown below.
This is a bug reported to the upstream OpenCV repository.
Try out yourself:
$ wget https://github.com/nglee/opencv_test/releases/download/hacking-opencv/hacking-opencv.tar.gz
$ tar xzf hacking-opencv.tar.gz
$ cd hacking-opencv
$ make
$ ./opencv_hack
The OpenCV 3.4.1 library used in the source was built on Ubuntu 16.04:
$ wget https://github.com/opencv/opencv/archive/3.4.1.tar.gz
$ tar -xzf 3.4.1.tar.gz
$ cd opencv-3.4.1
$ mkdir build/install -p && cd build
$ cmake ../ -DCMAKE_INSTALL_PREFIX=/home/nglee/Workspace/opencv-3.4.1/build/install -DWITH_CUDA=on -DCUDA_ARCH_BIN=5.0 -DBUILD_JAVA=off -DBUILD_PERF_TESTS=off -DBUILD_TESTS=off -DBUILD_PACKAGE=off -DBUILD_opencv_cudaarithm=off -DBUILD_opencv_cudabgsegm=off -DBUILD_opencv_cudacodec=off -DBUILD_opencv_cudafeatures2d=off -DBUILD_opencv_cudafilters=off -DBUILD_opencv_cudaimgproc=off -DBUILD_opencv_cudalegacy=off -DBUILD_opencv_cudastereo=off -DBUILD_opencv_cudawarping=off -DBUILD_opencv_java_bindings_generator=off -DBUILD_opencv_python_bindings_generator=off
$ make -j 7 && make install
The original test image is from Pixabay.