VisFactor benchmark shows frontier MLLMs reach at most 54% on human visual cognition subtests, failing on mental rotation, spatial relations, and figure-ground discrimination.
TRUE” if it shows the rela- tionships for the three groups, “FALSE
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CV 1years
2025 1verdicts
UNVERDICTED 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
Human Cognitive Benchmarks Reveal Foundational Visual Gaps in MLLMs
VisFactor benchmark shows frontier MLLMs reach at most 54% on human visual cognition subtests, failing on mental rotation, spatial relations, and figure-ground discrimination.