Unified-IO 2: Scaling Autoregressive Multimodal Models with Vision, Language, Audio, and Action

Benchmark Model Rank Results
visual-grounding-on-refcoco-valUnified-IO 2 XXL#3Accuracy: 90.7
science-question-answering-on-scienceqaUnified-IO 2 XL#8Avg. Accuracy: 88.8
object-hallucination-evaluation-on-popeUnified-IO 2 XXL#10F1: 87.7
visual-question-answering-on-ok-vqaUnified-IO 2 XXL#12Accuracy: 55.5
audio-classification-on-vggsoundUnified-IO 2 XXL#17Top 1 Accuracy: 47.7
question-answering-on-boolqUnified-IO 2 XXL#50Accuracy: 77.3
action-classification-on-kinetics-400Unified-IO 2 XXL#158Acc@1: 73.8
sentence-completion-on-hellaswagUnified-IO 2 XXL (zero-shot)#167Accuracy: 54.3