search

TUM-CAMP-NARVIS / nvenc_rtsp

NvPipe with RTSP server/client
Other
6 stars 3 forks source link

Need Windows (VS2017) compatibility #3

Open ulricheck opened 5 years ago

ulricheck commented 5 years ago

It turns out, that we will need to run our compute nodes initially on windows 10 (using VS2017) as our main compiler.

I've tried to adapt the nvenc_rtsp code to compile on windows, but could not resolv all issues due to time constraints.

I've added a (win-only) dependency in libunistd (https://github.com/robinrowe/libunistd) in the hope to get sufficient posix compatibility out of it, but it chokes on several string operations.

Resolution:

kyuq commented 5 years ago

I fixed the compile errors, but I were not able to test it, since opencv didn't want to work for me via conan in windows. However, the changes compile in branch conan_free_version in windows.

Can you try compiling it on your machine again and see if the changes solved your problem?

ulricheck commented 5 years ago

I've fixed some issues with linking (visibility of classes) of shared libs on windows. Now everything compiles (incl. device_comm_videostream) on windows and linux.

however, on both platforms only sending (verified with VLC) works, but receiving does not (segfault on windows, no error but blank screen on linux). Settings were default (H264, lossless, 30fps, 30mbit)

please use the artekmed_release_tools from: https://github.com/TUM-CAMP-NARVIS/artekmed_release_tools to install a complete set of libraries (maybe clean your conan cache beforehand) and try to find out why receiving is not working at all. you can use the windows_default_build.yml to build on windows and ubuntu_default_build.yml to build on linux ubuntu 18.04.

kyuq commented 5 years ago

On Linux, it worked for me for latest commit (c197300). Did you try to force-rebuild opencv? While the screen is blank, can you confirm the camera is running? (For instance, some cameras a have a small indicator LED)

I will checkout the segfault for windows these days

ulricheck commented 5 years ago

Ok - i think i found a problem - it seems that the library cannot handle larger frames (kinect4azure is capturing with 1536P) .. i've created a testbed environment for the nvenc_streaming with k4a (and an opencv demo as well) at: https://github.com/ulricheck/artekmed_k4a_capture

please install and compile:

mkdir cmake-build-debug
conan install .. --build missing
cmake ..
make

run stream sender:

source activate_run.sh
bin/opencv_stream_sender --scale 2

you can increase the scale factor (each time the image size is doubled) until it fails.

run stream receiver:

source activate_run.sh
bin/k4a_stream_receiver --nodepth --debug