List your company

Audio-visual event recognition on MIT-MUSIC (test)

91.35AV AccuracyMinSim
58.350866.917975.48584.0521Apr 5, 2021

Evaluation results

MethodLinks
MinSimDefense=Dissimilarity constraint (proposed)2021.0491.3570.1946.1536.5450.9616.35
MinSim+ExFMemDefense=Full defense model (proposed)2021.0490.3873.0853.8542.3156.4120.83
MinSim + ExFMem (Full)Attacker=MIM, Attack Strength=0.062021.0490.3864.4227.8818.2736.8626.6
MaxSimDefense=Maximizing audio-visual similarity2021.0489.4252.8845.1931.7343.276.73
ExFMemDefense=External feature memory banks (proposed)2021.0489.4253.855020.1941.344.8
MaxSimAttacker=MIM, Attack Strength=0.062021.0489.422531.7315.3824.0412.84
NoneDefense=No defense baseline2021.0488.4651.9245.1915.3837.50
NoneAttacker=MIM, Attack Strength=0.062021.0488.4620.1916.35012.180
PCLDefense=Prototype conformity loss2021.0483.6581.7337.536.5451.919.6
PCLAttacker=MIM, Attack Strength=0.062021.0483.6579.8117.3117.3138.1421.15
Unimodal VDefense=Visual-only network2021.0481.7381.7311.5411.5434.94-9.29
Unimodal VAttacker=MIM, Attack Strength=0.062021.0481.7381.7311.5411.5434.9416.03
Unimodal ADefense=Audio-only network2021.0459.62059.62019.87-46.47
Unimodal AAttacker=MIM, Attack Strength=0.062021.0459.62059.62019.87-21.15