List your company
ResearchFormerly WizWand

Knowledge-Based Visual Question Answering on OK-VQA v1.0 (test)

61.2AccuracyOurs + MCAN
37.43643.605549.77555.9445Oct 20, 2023
Updated 8mo ago

Evaluation results

MethodLinks
Ours + MCANKnowledge Resources=Frozen LLaMA 2 (13B)2023.1061.2
ProphetKnowledge Resources=Frozen GPT-3 (175B)2023.1061.1
Ours + MCANKnowledge Resources=Frozen LLaMA (13B)2023.1060.02
OursKnowledge Resources=Frozen LLaMA 2 (13B)2023.1059.07
OursKnowledge Resources=Frozen LLaMA (13B)2023.1058.69
REVIVE_large (ensemble)Knowledge Resources=Wikidata+Frozen GPT-3 (175B)2023.1058
Flamingo (80B)Knowledge Resources=Multimodal Pretraining2023.1057.8
REVIVE_large (single)Knowledge Resources=Wikidata+Frozen GPT-3 (175B)2023.1056.6
KAT_large (ensemble)Knowledge Resources=Wikidata+Frozen GPT-3 (175B)2023.1054.41
Unified-IO (2.8B)Knowledge Resources=Multimodal Pretraining2023.1054
KAT_large (single)Knowledge Resources=Wikidata+Frozen GPT-3 (175B)2023.1053.09
KAT_base (single)Knowledge Resources=Wikidata+Frozen GPT-3 (175B)2023.1050.58
PICa-FullKnowledge Resources=Frozen GPT-3 (175B)2023.1048
MAVEXKnowledge Resources=Wikipedia+ConceptNet+Google Images2023.1039.4
KRISPKnowledge Resources=Wikipedia+ConceptNet2023.1038.35