Rethinking UMM Visual Generation: Masked Modeling for Efficient Image-Only Pre-training

Open paper
Benchmark Model Rank Results
text-to-image-generation-on-wiseIOMM-B 512#12Overall: 0.55
text-to-image-generation-on-wiseIOMM-L 512#15Overall: 0.53
text-to-image-generation-on-wiseIOMM-B 1024#17Overall: 0.50
text-to-image-generation-on-wiseIOMM-L 1024#19Overall: 0.48
image-editing-on-imgedit-dataIOMM-B (image-only pre-trained)#29Overall: 2.82Add: 3.84Adjust: 2.37Extract: 2.12Replace: 2.60…
image-editing-on-imgedit-dataIOMM-B (text-image pair pre-trained)#31Overall: 2.61Add: 3.18Adjust: 2.17Extract: 1.92Replace: 2.70…
text-to-image-generation-on-genevalIOMM-B 512#30Overall: 0.89
text-to-image-generation-on-genevalIOMM-B 1024#48Overall: 0.87
text-to-image-generation-on-genevalIOMM-L 512#49Overall: 0.87
text-to-image-generation-on-genevalIOMM-L 1024#66Overall: 0.85
image-generation-on-wiseIOMM-B 512#36Overall: 0.55
image-generation-on-wiseIOMM-L 512#40Overall: 0.53
image-generation-on-wiseIOMM-B 1024#42Overall: 0.50
image-generation-on-wiseIOMM-L 1024#44Overall: 0.48

Rank counts only results with a code link.