weikaih/imaginative-perception-token-pet-eval-ai2thor
Viewer • Updated • 278 • 51
Human-verified spatial mental modeling benchmarks: perspective taking, multiview counting, path tracing (AI2-THOR / Habitat / Real).
Note PET AI2-THOR - human-verified, in-domain
Note MVC AI2-THOR - human-verified, in-domain
Note PET Habitat - human-verified, different env
Note PT AI2-THOR - human-verified (td_ego_dir/td_path/td_path_arrow)
Note PT Real indoor - human-verified (td_path/td_path_arrow)
Note IPT MVC Mixed model (answer-only + imaginative inference)