MME Perception Evaluation

Offline

MME Perception Evaluation

Multi-modal Model Evaluation â€" 14 categories (10 perception + 4 cognition). Official scoring: accuracyÃ--100 + plus_scoreÃ--100 per category.

No VP runs found.

Launch evaluation â†'

Perception Tasks

CategoryBaselineRevision OnlyFixed GateFull MITCF
existenceâ€"â€"â€"â€"
countâ€"â€"â€"â€"
positionâ€"â€"â€"â€"
colorâ€"â€"â€"â€"
postersâ€"â€"â€"â€"
celebrityâ€"â€"â€"â€"
sceneâ€"â€"â€"â€"
landmarkâ€"â€"â€"â€"
artworkâ€"â€"â€"â€"
OCRâ€"â€"â€"â€"

Cognition Tasks

CategoryBaselineRevision OnlyFixed GateFull MITCF
commonsense reasoningâ€"â€"â€"â€"
numerical calculationâ€"â€"â€"â€"
text translationâ€"â€"â€"â€"
code reasoningâ€"â€"â€"â€"

MME Scoring

Official MME score per category = accuracyÃ--100 + plus_scoreÃ--100, where plus_score credits image pairs where both questions are answered correctly. Total score is the sum across all categories (max ~3,000 for 14 categories at perfect accuracy with all pairs correct).