List your company

Visual Question Answering on VQA 2.0 (test)

80.2AccuracyPretrained FT SOTA
27.47241.16154.8568.539Jul 10, 2018Mar 6, 2019Oct 31, 2019Jun 26, 2020Feb 20, 2021Oct 17, 2021Jun 13, 2022

Evaluation results

MethodLinks
Pretrained FT SOTAFT=true, Shot=10K2022.0480.2—
FlamingoFT=false, Shot=322022.0467.6—
XNMsexpert layout=no2018.1267.5—
Up-Downexpert layout=no2018.1266.3—
StackNMNexpert layout=no, reported_in_original_paper=true2018.1264.1—
N2NMNexpert layout=yes, reported_in_original_paper=true2018.1263.3—
FlamingoFT=false, Shot=42022.0463.1—
Flamingo-9BFT=false, Shot=322022.0460.4—
Marginal LikelihoodObjective=log E[p], Expectation calculation=Enum2018.0760.331.69
Soft AttentionObjective=log p(y | E[z]), Expectation calculation=N/A2018.0758.931.76
Variational AttentionObjective=Eq [log p] - KL, Expectation calculation=Enum2018.0758.441.69
Hard AttentionObjective=Ep [log p], Expectation calculation=Enum2018.0757.61.78
Variational AttentionObjective=Eq [log p] - KL, Expectation calculation=Sample2018.0757.521.75
Flamingo-3BFT=false, Shot=322022.0457.1—
Hard AttentionObjective=Ep [log p], Expectation calculation=Sample2018.0756.31.82
Flamingo-9BFT=false, Shot=42022.0456.3—
FlamingoFT=false, Shot=02022.0456.3—
Flamingo-3BFT=false, Shot=42022.0453.2—
Flamingo-9BFT=false, Shot=02022.0451.8—
Flamingo-3BFT=false, Shot=02022.0449.2—
METALMevaluation protocol=zero-shot2022.0641.1—
VLKD_ViT-B/16evaluation protocol=zero-shot2022.0638.6—
Zero/Few shot SOTAFT=false, Shot=162022.0438.2—
Frozenevaluation protocol=zero-shot2022.0629.5—