Retrieval-Augmented Multimodal Language Modeling

Benchmark Model Rank Results
image-captioning-on-cocoDALL-E–CIDEr: 20.2
image-captioning-on-cocoFlamingo (3B; 4-shot)–CIDEr: 85
image-captioning-on-cocoFlamingo (80B; 4-shot)–CIDEr: 103
image-captioning-on-cocoParti–CIDEr: 83.9
image-captioning-on-cocoRA-CM3 (2.7B)–CIDEr: 89.1
image-captioning-on-cocoVanilla CM3–CIDEr: 71.9
image-captioning-on-cocoX-LXMERT–CIDEr: 55.8
image-captioning-on-cocominDALL-E–CIDEr: 48
image-captioning-on-cocoruDALL-E-XL–CIDEr: 38.7
text-to-image-generation-on-cocoDALL-E (12B)–FID: 28
text-to-image-generation-on-cocoRA-CM3 (2.7B)–FID: 15.7
text-to-image-generation-on-cocoStable Diffusion–FID: 12.63
text-to-image-generation-on-cocoVanilla CM3–FID: 29.5