Open-vocabulary sensor-language retrieval on ActivityNarrated XP 1.0 (Cross-Position)
34R@1Spectral + Q-Former + T.a.
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Spectral + Q-Former + T.a.Discretization=Spectral, Aggregation=Q-Former, T.a.=true2026.04 | 34 | 66.3 | 56 | 64 | |
| Spectral + Q-FormerDiscretization=Spectral, Aggregation=Q-Former2026.04 | 30.9 | 61.5 | 50 | 59 | |
| Spectral + attentionDiscretization=Spectral, Aggregation=attention2026.04 | 28 | 54.1 | 43 | 55 | |
| Spectral + mean poolingDiscretization=Spectral, Aggregation=mean pooling2026.04 | 25.7 | 51.5 | 41 | 53 | |
| Raw + Q-FormerDiscretization=Raw, Aggregation=Q-Former2026.04 | 23.2 | 50 | 39 | 51 | |
| Raw + attentionDiscretization=Raw, Aggregation=attention2026.04 | 21 | 46.5 | 38 | 49 | |
| PaLM-E2026.04 | 20.9 | 44.2 | 36 | 47 | |
| Raw + mean poolingDiscretization=Raw, Aggregation=mean pooling2026.04 | 20.1 | 45.8 | 33 | 47 | |
| OVHAR2026.04 | 19.3 | 44.2 | 32 | 46 | |
| IMU2CLIP2026.04 | 8.9 | 29.5 | 19 | 30 |