`renderObjectImage` slows the `step` function dramatically - can it be its own action?

From quick timings on a few machines, step with movement actions takes .1 to .15 seconds when renderObjectImage=False. This time per step jumps up to .8-.9 seconds with renderObjectImage=True. In some applications, we don’t need the object segmentation mask at every timestep. Would it be possible to make renderObjectImage an action instead, such that we can take a step to render and get the object segmentation frame (a slow step), but have the rest of the steps not bothering generating it so that the simulator runs faster?

[ekolve]

This is now fixed with 1a2bdf51801e72bcc39366d6af00fb801cc5b2ec and 1a2bdf51801e72bcc39366d6af00fb801cc5b2ec

[mattdeitke]

Hey Jesse, this is supported, but not currently documented, with:

from ai2thor.controller import Controller
c = Controller()
event = c.step(action='MoveAhead', renderObjectImage=True)
event.instance_segmentation_frame  # not empty