Sign recognition on BSL-1K 37K_Rec (test)
65Per-Instance Top-1 AccM+D+A
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| M+D+ATraining Labels=Mouthing (M) + Dictionary (D) + Attention (A), Number of Annotations=786K, Architecture=4-layer MLP2021.03 | 65 | 82.6 | 37.9 | 56.3 | |
| D+ATraining Labels=Dictionary (D) + Attention (A), Number of Annotations=276K, Architecture=4-layer MLP2021.03 | 64.2 | 81.7 | 36 | 54 | |
| M+DTraining Labels=Mouthing (M) + Dictionary (D), Number of Annotations=678K, Architecture=4-layer MLP2021.03 | 63.5 | 82.1 | 37.2 | 56.4 | |
| DTraining Labels=Dictionary (D), Number of Annotations=510K, Architecture=4-layer MLP2021.03 | 62.1 | 80.8 | 35.1 | 54.3 | |
| M+ATraining Labels=Mouthing (M) + Attention (A), Number of Annotations=276K, Architecture=4-layer MLP2021.03 | 58.5 | 75.5 | 30.4 | 45.9 | |
| ATraining Labels=Attention (A) spottings, Number of Annotations=107K, Architecture=4-layer MLP2021.03 | 54 | 67.9 | 24.8 | 35.5 | |
| MTraining Labels=Mouthing (M), Number of Annotations=169K, Architecture=4-layer MLP2021.03 | 40.8 | 62.2 | 21.7 | 38.5 |