List your company
ResearchFormerly WizWand

Visual Question Answering on VQAv2 (val)

75.33Accuracy (Overall)Shikra
11.026827.720944.41561.1091Oct 17, 2022Dec 26, 2022Mar 6, 2023May 15, 2023Jul 24, 2023Oct 2, 2023Dec 11, 2023
Updated 8mo ago

Evaluation results

MethodLinks
ShikraModel Size=13B2023.0675.33————
BLIP-22023.0665.2————
VPGTrans2023.0665.2————
PNP-VQA11BLanguage Model=UnifiedQAv2, Language #Params=11.3B, Language VL-aware=false, Vision Model=BLIP-Caption, Vision #Params=446M, Vision VL-aware=true, Zero-shot=true2022.1063.3————
PNP-VQA3BLanguage Model=UnifiedQAv2, Language #Params=2.9B, Language VL-aware=false, Vision Model=BLIP-Caption, Vision #Params=446M, Vision VL-aware=true, Zero-shot=true2022.1062.1————
PNP-VQAlargeLanguage Model=UnifiedQAv2, Language #Params=738M, Language VL-aware=false, Vision Model=BLIP-Caption, Vision #Params=446M, Vision VL-aware=true, Zero-shot=true2022.1057.5————
PNP-VQAbaseLanguage Model=UnifiedQAv2, Language #Params=223M, Language VL-aware=false, Vision Model=BLIP-Caption, Vision #Params=446M, Vision VL-aware=true, Zero-shot=true2022.1054.3————
BLIP-2+ (OPT2.7B)Backbone=OPT-2.7B, #Pretrain Image-Text=129M, Pretrain Uni-Text=false, Evaluation Protocol=Zero-shot2023.0753.5————
P-Former (OPT2.7B)Backbone=OPT-2.7B, #Pretrain Image-Text=4M, Pretrain Uni-Text=true, Evaluation Protocol=Zero-shot2023.0752.6————
BLIP-2 + MAFAEvaluation Protocol=Zero-shot2023.1250.8————
FewVLMlargeLanguage Model=T5, Language #Params=740M, Language VL-aware=false, Vision Model=Faster R-CNN, Vision #Params=64M, Vision VL-aware=false, Zero-shot=true2022.1047.7————
FewVLM#Pretrain Image-Text=9.2M, Pretrain Uni-Text=false, Evaluation Protocol=Zero-shot2023.0747.7————
BLIP-2 (OPT2.7B)Backbone=OPT-2.7B, #Pretrain Image-Text=4M, Pretrain Uni-Text=false, Evaluation Protocol=Zero-shot2023.0746.8————
BLIP-2Evaluation Protocol=Zero-shot2023.1246.6————
Few VLMbaseLanguage Model=T5, Language #Params=224M, Language VL-aware=false, Vision Model=Faster R-CNN, Vision #Params=64M, Vision VL-aware=false, Zero-shot=true2022.1043.4————
VLKD ViT-L/14Language Model=BART, Language #Params=408M, Language VL-aware=true, Vision Model=ViT-L/14, Vision #Params=305M, Vision VL-aware=true, Zero-shot=true2022.1042.6————
VLKD#Pretrain Image-Text=3M, Pretrain Uni-Text=false, Evaluation Protocol=Zero-shot2023.0742.6————
VLKD ViT-B/16Language Model=BART, Language #Params=407M, Language VL-aware=true, Vision Model=ViT-B/16, Vision #Params=87M, Vision VL-aware=true, Zero-shot=true2022.1038.6————
Frozen#Pretrain Image-Text=3M, Pretrain Uni-Text=false, Evaluation Protocol=Zero-shot2023.0729.6————
FrozenLanguage Model=GPT-like, Language #Params=7B, Language VL-aware=false, Vision Model=NF-ResNet-50, Vision #Params=40M, Vision VL-aware=false, Zero-shot=true2022.1029.5————
VL-T5no-vqaLanguage Model=T5, Language #Params=224M, Language VL-aware=false, Vision Model=Faster R-CNN, Vision #Params=64M, Vision VL-aware=false, Zero-shot=true2022.1013.5————
Frozenuses VQAv2 training data=false2021.06—29.535.738.2—
Frozen finetuneduses VQAv2 training data=false2021.06—2428.229.2—
Frozen scratchuses VQAv2 training data=false2021.06—000—
Frozen train-blinduses VQAv2 training data=false2021.06—26.233.533.3—
Frozen VQAuses VQAv2 training data=true2021.06—48.4———
Frozen VQA-blinduses VQAv2 training data=true2021.06—39.1———
Oscaruses VQAv2 training data=true2021.06—73.8———