Bug Running ml-agents with visual observations on headless server with xvfb-run causes repeated frames

When I run my experiment locally it works perfectly and I am available to get a smooth movie of what my agent sees, but when I run it on a headless server with xvfb-run, the visual observations get repeated for 6-8 frames so that the movie that I generate is frozen for 6-8 frames at a time.

I know this is somewhat a vague description and it is with a custom environment so I won't provide reproduction steps but I was just wondering if anyone had ever encountered this bug or might have some idea what might be causing it.

The xvfb command I am using is `xvfb-run --auto-servernum --server-args='-screen 0 640x480x24:32' python run.py -batchmode`

(run.py is where I initialize the environment, collect the observations from the environment and turn them into a movie)

Thanks!