Preference
MJ-Bench released judge observations
Original released multimodal judge preferences and bias ratings, linked to unchanged MJ-Bench image inputs.
15,630items
56subjects
MJ-Bench and ImageReward HF cards declare Apache-2.0. Author code/result repositories declare MIT. The captured HPDv2 and Pick-a-Pic mirror cards contain no explicit dataset license; retain their source attribution without extending the other releases' license declarations.license
reward_modelingdomain
preferencedomain
imagemodality
textmodality
item-level responses released
Saturation status: No
Response matrix
Loading response matrix…
Correct (1)Incorrect (0)Unobserved
Scale: 1 = correct · 0 = incorrect