Egocentric retrieval on Ego4D NLQ 5% fraction (val)
20.01Recall@1 (R1)DenoiseLoc
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| DenoiseLocevaluation protocol=supervised2023.05 | 20.01 | 25.76 | 34.77 | 24.99 | 30.82 | 39.47 | 27.85 | 32.94 | 42.13 | |
| Ego-NLSOM (d.) w. exec.# of agents=4, Rounds=1, post-processing=w. exec., strategy=democracy, evaluation protocol=zero-shot2023.05 | 9.38 | 10.42 | 17.19 | 17.71 | 18.23 | 29.17 | 22.92 | 25 | 39.58 | |
| Ego-NLSOM (d.)# of agents=4, Rounds=1, strategy=democracy, evaluation protocol=zero-shot2023.05 | 8.85 | 9.35 | 14.58 | 16.67 | 16.67 | 23.96 | 18.75 | 20.31 | 28.12 | |
| Ego-NLSOM (d.) w. exec.# of agents=4, Rounds=2, post-processing=w. exec., strategy=democracy, evaluation protocol=zero-shot2023.05 | 8.85 | 11.98 | 16.67 | 16.15 | 18.75 | 29.18 | 23.44 | 27.05 | 40.1 | |
| Ego-NLSOM (m.) w. exec.# of agents=4, Rounds=1, post-processing=w. exec., strategy=monarchy, evaluation protocol=zero-shot2023.05 | 7.81 | 9.9 | 13.02 | 16.67 | 18.75 | 29.17 | 25 | 27.08 | 40.62 | |
| Ego-NLSOM (m.) w. exec.# of agents=4, Rounds=2, post-processing=w. exec., strategy=monarchy, evaluation protocol=zero-shot2023.05 | 7.14 | 8.79 | 13.19 | 14.29 | 17.03 | 24.18 | 23.63 | 25.82 | 37.36 | |
| Ego-NLSOM (m.)# of agents=4, Rounds=1, strategy=monarchy, evaluation protocol=zero-shot2023.05 | 6.77 | 8.33 | 9.9 | 14.58 | 16.15 | 21.88 | 19.27 | 20.83 | 25.52 | |
| individual# of agents=1, Rounds=N/A, evaluation protocol=zero-shot2023.05 | 5.21 | 6.77 | 9.38 | 9.38 | 10.94 | 15.1 | 10.42 | 11.98 | 16.15 | |
| random# of agents=0, Rounds=N/A, evaluation protocol=zero-shot2023.05 | 4.69 | 5.21 | 12.5 | 7.29 | 7.81 | 20.31 | 10.42 | 11.46 | 30.21 | |
| grid# of agents=0, Rounds=N/A, evaluation protocol=zero-shot2023.05 | 3.12 | 3.65 | 7.81 | 6.25 | 7.81 | 15.1 | 11.98 | 14.06 | 27.6 |