MultiStAR decomposes RAVEN abstract reasoning into five staged sub-tasks and MSEval scores partial progress, showing MLLMs handle basic perception but fail at rule deduction.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Beyond Perception: Evaluating Abstract Visual Reasoning through Multi-Stage Task
MultiStAR decomposes RAVEN abstract reasoning into five staged sub-tasks and MSEval scores partial progress, showing MLLMs handle basic perception but fail at rule deduction.