Debug flag affects agent input and behavior

[AdamGleave]

Specifying the debug flag changes not just the visualization, but also the camera input from the agent; see attached screen capture. This substantially changes the behavior of agents that depend on RGB input.

This should at the least be documented, but ideally there would be a way to fix this as it makes the debugging flags much less useful.

In branch carla_challenge, I executed the imitation learning baseline test three times with and without the debug flag: CHALLENGE_PHASE_CODENAME=dev_track_2 python3 ${ROOT_SCENARIO_RUNNER}/srunner/challenge/challenge_evaluator_routes.py --scenarios=${ROOT_SCENARIO_RUNNER}/srunner/challenge/all_towns_traffic_scenarios1_3_4.json --routes=${ROOT_SCENARIO_RUNNER}/srunner/challenge/routes_debug.xml --agent=../coiltraine/drive/CoILBaseline.py --config=../coiltraine/drive/sample_agent.json --debug=0 --filename=debug0-rep2.json

With debug=0, I got average route points / infraction points of 12.73 / 17.5 (rep 0), 12.82 / 15.0 (rep 1) and 6.23 / 15.0 (rep 2). With debug=1, I got 4.02 / 17.5 (rep 0), 15.95 / 30.0 (rep 1) and 14.79 / 25.0 (rep 2).

Raw results below.

debug0-rep0.log debug0-rep1.log debug0-rep2.log debug1-rep0.log debug1-rep1.log debug1-rep2.log

[RuihanGao]

Same problem. The agent does not move at all with the debug flag. Please kindly help. Thanks in advance.