EGL init failed

[HX-Tfd]

I'm running the preprocessing script on a remote machine with headless rendering, and I get outputs like this:

Unable to read texture 'texture_7'
Unable to read texture 'texture_8'
EGL init failed
EGL init failed

resulting in no .npz files produced in the data folder.

As far as I understand, the textures are not needed for sampling, so we can safely ignore the warning. But I have no idea where the EGL issue comes from. I have the following modules loaded on the machine:

  1) StdEnv      3) libpng/1.6.34   5) mesa/18.3.6    7) nanoflann/1.2.3   9) python/3.8.5  11) glew/2.0.0  13) libtiff/4.0.10  15) lz4/1.9.3
  2) gcc/8.2.0   4) eigen/3.3.9     6) cmake/3.20.3   8) openblas/0.2.20  10) cuda/11.2.2   12) libjpeg/9d  14) ffmpeg/3.2.4    16) zstd/1.3.0

Could pls someone help? Thanks!

[gongxiaoran]

Hi, I also encountered this problem. Have you solved it?

[thinhngo-x]

Hi, I also encountered the same issue when running the code on the remote server with X11 enabled. I finally manged to solve this by following instructions here.

Further explanation is that your remote server has linked to the incorrect EGL library. Thus, the solution is to:

1. Find your nvidia EGL/GLX/OpenGL library (*.so files) in /usr/lib.
2. Pre-load the *.so files.

For my remote machine, I managed to correct the issue by putting all these into a bash file and run it:

export PANGOLIN_WINDOW_URI=headless://
export LD_PRELOAD="/usr/lib/x86_64-linux-gnu/libEGL.so"  # pre-load EGL
export MESA_GL_VERSION_OVERRIDE=3.3  # correct GLSL version
python preprocess_data.py ...