Audio-visual event recognition on Kinetics-Sounds (test)
72.71AV ScoreExFMem
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| ExFMemDefense=External feature memory banks (proposed)2021.04 | 72.71 | 41.56 | 29.93 | 10.44 | 27.31 | 3.99 | |
| NoneDefense=No defense baseline2021.04 | 72.42 | 36.4 | 26.35 | 8.09 | 23.61 | 0 | |
| NoneAttacker=MIM, Attack Strength=0.062021.04 | 72.42 | 11.18 | 9.63 | 32 | 7.04 | 0 | |
| MaxSimDefense=Maximizing audio-visual similarity2021.04 | 71.39 | 34.95 | 29.57 | 21.46 | 28.66 | 4.02 | |
| MaxSimAttacker=MIM, Attack Strength=0.062021.04 | 71.39 | 17.65 | 18.78 | 1,389 | 16.77 | 8.7 | |
| MinSim+ExFMemDefense=Full defense model (proposed)2021.04 | 71.33 | 55.96 | 30.57 | 24.9 | 37.14 | 12.44 | |
| MinSim + ExFMem (Full)Attacker=MIM, Attack Strength=0.062021.04 | 71.33 | 44.72 | 16.04 | 999 | 23.58 | 15.45 | |
| MinSimDefense=Dissimilarity constraint (proposed)2021.04 | 70.88 | 52.42 | 28.12 | 21.62 | 34.05 | 8.99 | |
| Unimodal VDefense=Visual-only network2021.04 | 66.08 | 66.08 | 18.72 | 18.72 | 34.5 | 4.55 | |
| Unimodal VAttacker=MIM, Attack Strength=0.062021.04 | 66.08 | 66.08 | 5.77 | 577 | 25.87 | 12.49 | |
| PCLDefense=Prototype conformity loss2021.04 | 64.5 | 63.43 | 29.28 | 28.67 | 40.46 | 8.93 | |
| PCLAttacker=MIM, Attack Strength=0.062021.04 | 64.5 | 62.98 | 20.26 | 1,997 | 34.4 | 19.44 | |
| Unimodal ADefense=Audio-only network2021.04 | 35.99 | 1.87 | 35.99 | 1.87 | 13.24 | -46.8 | |
| Unimodal AAttacker=MIM, Attack Strength=0.062021.04 | 35.99 | 0.1 | 35.99 | 10 | 12.06 | -31.41 |