Gflags error when running [roslaunch khronos_ros uhumans2_khronos.launch]

[18lalala]

Hi,
Glad to see your fantastic work. I succeeded to compile the project and wanted to run a demo by running roslaunch khronos_ros uhumans2_khronos.launch and I met these errors.

ERROR: unknown command line flag 'minloglevel'
ERROR: unknown command line flag 'v'
ERROR: unknown command line flag 'alsologtostderr'
ERROR: unknown command line flag 'colorlogtostderr'
ERROR: unknown command line flag 'v'

I found the reason maybe the function google:ParseCommandLineFlags(&argc, &argv, true) can't parse the flags defined in gflags, but I can not fix it. Do you have any ideas? Thanks a lot.

here are more information about my ros enviroment
ubuntu 20.04
ros-noetic
libgoogle-glog-dev=0.4.0-1build1
libgflags-dev=2.2.2-1build1

[yunzc]

My guess is that khronos and glog might have conflicting gflags versions.

What's the output of ldd <catkin-ws>/lib/libkhronos.so|grep glog and ldd <catkin-ws>/lib/libkhronos.so|grep gflag for you?

[18lalala]

Thanks, the glog version is 0.4.0, and gflags 2.2.2
I met this error on NVIDIA JETSON ORIN, while it did not appear on a x86 computer. I have checked that versions of glog and gflags are the same.

[yunzc]

Sorry, to clarify, I am suspecting that you might have duplicated glog installs on your system, hence was curious to see your ldd output and exactly what is being used. Local or source installs of glog take priority over the system install and are built without gflags support by default, which would be an explanation to the error that you are seeing.

Regarding Jetson vs x86: to the best of our knowledge, the system glog package doesn't change that option between architectures.

You could also try and build glog from source and install it with gflags support turned on and build against that instead

[18lalala]

Thanks a lot. I will check the installation of glog. BTW, will you update the version with VIO and Semantic Segmentation Module via raw RGB-D camera input? I added a semseg model which is trained on ADE20K and outputs a label from 150 classes to just replace the gt labels, and got bad performance, so looking forward for your real world version.

[Schmluk]

Quick updated: We've (finally) released the real world datasets and open-set segmentation tools. Please feel free to give them a look!