Will's profile - activity

Enabling Address Sanitizer for OpenCV builds Is there a way to enable Address Sanitizer for OpenCV builds? I've tried a

Enabling Address Sanitizer for OpenCV builds Is there a way to enable Address Sanitizer for OpenCV builds? I've tried a

opencv_dnn provides incorrect inferences after transform_graph I am running into a problem that has been encountered by

Unfortunately, OpenCV currently needs libtiff4, which Ubuntu 14.04 has dropped from their packaging system. As a workaround, specify -DBUILD_TIFF=ON to CMake when configuring OpenCV. This will have OpenCV build the distributed version of libtiff4, and should solve your linker problems.

Hope that helps!

@glopes Thanks for the info! Glad to know I'm not the only one experiencing the performance drop. I will definitely give your work around a shot. I'll also start gathering info to submit an FFmpeg bug report to hopefully fix the upstream problem.

@jestshen Thanks for the idea, but unfortunately both my software and the FFmpeg library I built have optimizations enabled.

I have been using OpenCV's VideoCapture class to decode several H.264 camera streams simultaneously in a custom video viewer application. Unfortunately, when I upgraded OpenCV 2.4.6 to 2.4.9 my client reported a large slow down in the framerate at which the viewer shows frames. The framerates are roughly half of what they used to be (i.e., 30fps is now 15fps viewer speed).

I have since built FFmpeg from source myself, and am seeing very similar slow downs with my own build of FFmpeg 2.3.3 (as well as git head) (vs. 2.0.2, which is present in the OpenCV 2.4.9) using OpenCV's FFmpeg wrapper implementation.

EDIT: Both my FFmpeg build and program are built in Release mode with optimizations.

I was wondering if anyone on the OpenCV team happened to have custom build flags for 2.4.6, or if they are the same as is shown in the GitHub readme? Was libx264 possibly included in some of the previous builds; I understand that libx264 is a higher performance H.264 codec, so maybe that accounted for the speed decrease? Has anyone else noticed this slow down? It may not occur when reading a single stream, but it is definitely noticeable when decoding many. This may turn into an FFmpeg bug report, but just wanted to make sure I wasn't seeing things :D

I have been able to, in the past, set the compiler options via the command-line like this:

CXXFLAGS="-O2 -mcpu=cortex-a8 -mfpu=neon -mfloat-abi=softfp" CFLAGS="-O2 -mcpu=cortex-a8 -mfpu=neon -mfloat-abi=softfp" cmake -DCMAKE_TOOLCHAIN_FILE=~/src/arm-toolchain.cmake ..

However, as of OpenCV 2.4.3 the CMake scripts seem to be actively ignoring/resetting the user specified CFLAGS. Is there a way around this I'm not seeing, or should I file a bug report?

Thanks for the help!