HPSv3
A human preference score reward model used to evaluate the quality of generated images. It is trained on human preference data to assign quality scores to images produced by generative models.
Timeline (last 90 days)
GraftOOD causes HPSv3 to slightly decrease from 8.10 to 7.32, while VisionReward remains essentially unchanged.
On the Pick-a-Pic dataset using DreamSim as the diversity metric, with FLUX.1-dev as the base model and HPSv3 as the reward model, SatisDive improved the worst candidate reward by up to 0.43 compared to FK steering.
Both HPSv3 and ImageReward are human preference reward models trained on large-scale human-annotated image preference datasets to predict human satisfaction scores for generated images.
All Facts (3)
GraftOOD causes HPSv3 to slightly decrease from 8.10 to 7.32, while VisionReward remains essentially unchanged.
50%UnverifiedOn the Pick-a-Pic dataset using DreamSim as the diversity metric, with FLUX.1-dev as the base model and HPSv3 as the reward model, SatisDive improved the worst candidate reward by up to 0.43 compared to FK steering.
50%UnverifiedBoth HPSv3 and ImageReward are human preference reward models trained on large-scale human-annotated image preference datasets to predict human satisfaction scores for generated images.
50%