Imaginative Perception Token Data
Datasets for the Imaginative Perception Token (IPT) paper: MVC + PET + PT, training + human-verified eval.
Viewer • Updated • 20k • 1.75k • 1Note MVC IPT / Visual CoT training (17,079)
weikaih/imaginative-perception-token-mvc-textcot
Viewer • Updated • 16.8k • 238Note MVC Text CoT training (16,808)
weikaih/imaginative-perception-token-pet-ipt
Viewer • Updated • 20.5k • 1.35k • 1Note PET IPT / Visual CoT training (20,531)
weikaih/imaginative-perception-token-pet-textcot
Viewer • Updated • 20.5k • 167Note PET Text CoT training (20,531)
leo66666/messytable
Viewer • Updated • 5.58k • 35Note MVC MessyTable real-world OOD
leo66666/scannet_counting
Viewer • Updated • 540 • 44Note MVC ScanNet OOD
weikaih/imaginative-perception-token-pet-eval-ai2thor
Viewer • Updated • 278 • 1.13kNote PET AI2-THOR - human-verified, in-domain
luckychao/vlmevalkit_tsv
Preview • Updated • 197Note SAT (perspective subset) OOD
linjieli222/spatial-imaginative-token-pt-ipt
Viewer • Updated • 11.2k • 311Note PT (Path Tracing) train - IPT (11,204)
linjieli222/spatial-imaginative-token-pt-answeronly
Viewer • Updated • 11.2k • 265Note PT train - answer-only (serves no_thought + mixed) (11,204)
linjieli222/spatial-imaginative-token-pt-textcot
Viewer • Updated • 11.2k • 138Note PT train - Text CoT (11,204)
weikaih/imaginative-perception-token-mvc-answeronly
Viewer • Updated • 17.1k • 154Note MVC answer-only - serves label-only + Mixed (17,079)
weikaih/imaginative-perception-token-pet-answeronly
Viewer • Updated • 20.5k • 162Note PET answer-only - serves label-only + Mixed (20,531)
weikaih/imaginative-perception-token-mvc-eval-ai2thor
Viewer • Updated • 260 • 638Note MVC AI2-THOR - human-verified, in-domain
weikaih/imaginative-perception-token-pet-eval-habitat
Viewer • Updated • 719 • 890Note PET Habitat - human-verified, different env
weikaih/imaginative-perception-token-pt-eval-ai2thor
Viewer • Updated • 453 • 436Note PT AI2-THOR - human-verified (td_ego_dir/td_path/td_path_arrow)
weikaih/imaginative-perception-token-pt-eval-real
Viewer • Updated • 332 • 308Note PT Real indoor - human-verified (td_path/td_path_arrow)
-
Imaginative Perception Tokens Enhance Spatial Reasoning in Multimodal Language Models
Paper • 2606.03988 • Published • 127
weikaih/imaginative-perception-token-mvc-mixed
Image-Text-to-Text • 15B • Updated • 13Note IPT MVC Mixed model (answer-only + imaginative inference)