UniWorld-V1: High-Resolution Semantic Encoders for Unified Visual Understanding and Generation

Benchmark Model Rank Results
image-editing-on-imgedit-dataUniWorld-V1#1Overall: 3.26Add: 3.82Adjust: 3.64Extract: 2.27Replace: 3.47
image-generation-on-wiseUniWorld-V1#3Overall: 0.55Cultural: 0.53Time: 0.55Space: 0.73Biology: 0.45
on-wiseUniWorld-V1#1Overall: 0.55Cultural: 0.53Time: 0.55Space: 0.73Biology: 0.45
text-to-image-generation-on-genevalUniWorld-V1 (Rewrite)#2Overall: 0.84Single Obj.: 0.98Two Obj.: 0.93Color Attri.: 0.71
text-to-image-generation-on-genevalUniWorld-V1#6Overall: 0.80Single Obj.: 0.99Two Obj.: 0.93Color Attri.: 0.70