List your company
ResearchFormerly WizWand

Human Preference Evaluation on ImageReward (test)

0.675Preference AccuracyMPS
0.512760.554880.5970.63912May 23, 2024Sep 22, 2024Jan 22, 2025May 24, 2025Sep 23, 2025Jan 23, 2026May 25, 2026
Updated 5mo ago

Evaluation results

MethodLinks
MPSTrained on=MHP datasets2024.050.675
MPS2026.050.675
HPSv3Backbone=Qwen2VL-7B2026.020.6703
HPSv3 - 7B2026.050.668
MPSBackbone=CLIP2026.020.6637
HPS v2Pre-trained=true2024.050.657
HPSv22026.050.657
HPSv2Backbone=CLIP2026.020.6562
ImageRewardBackbone=CLIP2026.020.6515
ImageRewardPre-trained=true2024.050.651
ImageReward2026.050.651
DRMWeights=Pre-trained, Epoch=1, Size=5122026.050.641
UnifiedRewardBackbone=LLaVA-OV-7B2026.020.6382
PickScorePre-trained=true2024.050.629
DRMWeights=Pre-trained, Epoch=1, Size=2562026.050.629
PickScoreBackbone=CLIP2026.020.6273
DiNa-LRMBackbone=SD3.5-M-2B, Multi-noise inference scaling=true2026.020.6175
PickScore2026.050.616
HPSPre-trained=true2024.050.612
HPS2026.050.612
LRM-SDXLBackbone=SDXL2026.020.6035
DiNa-LRMBackbone=SD3.5-M-2B, Multi-noise inference scaling=false2026.020.6034
LRM-SD1.5Backbone=SD-1.52026.020.5917
UnifiedReward-ThinkBackbone=LLaVA-OV-7B2026.020.5854
HPSv3 - 2B2026.050.579
Aesthetic ScorePre-trained=true2024.050.574
Aesthetic Score Predictor2026.050.574
CLIP ViT-H/142026.050.571
CLIP scorePre-trained=true2024.050.543
DRMWeights=Random, Epoch=3, Size=2562026.050.537
DRMWeights=Random, Epoch=1, Size=2562026.050.524
DRMWeights=Random, Epoch=2, Size=2562026.050.519