In scene 5cdEh9F2hJL, VLFM detected a wall-mounted TV, navigated to it, and called STOP, but the episode failed with distance_to_goal=12.83.
The semantic annotations contain:
231: wall tv
353: monitor
354: monitor
363: wall tv
However, the ObjectNav episode only includes objects 353 and 354 as valid tv_monitor goals. I verified that the local files exactly match the official HM3D archives.
Is excluding wall tv from tv_monitor intentional? If so, how should an agent visually distinguish valid monitors from wall-mounted TVs?
In scene
5cdEh9F2hJL, VLFM detected a wall-mounted TV, navigated to it, and calledSTOP, but the episode failed withdistance_to_goal=12.83.The semantic annotations contain:
However, the ObjectNav episode only includes objects 353 and 354 as valid
tv_monitorgoals. I verified that the local files exactly match the official HM3D archives.Is excluding
wall tvfromtv_monitorintentional? If so, how should an agent visually distinguish valid monitors from wall-mounted TVs?