Language Models are General-Purpose Interfaces

Benchmark Model Rank Results
image-captioning-on-nocaps-valMetaLM#2CIDEr: 58.7SPICE: 8.6
image-captioning-on-flickr30k-captions-testMetaLM#3CIDEr: 43.3SPICE: 11.7
visual-question-answering-on-vqa-v2-valMetaLM#9Accuracy: 41.1
visual-question-answering-on-ok-vqaMetaLM#36Accuracy: 11.4