Egocentric Human Activity Recognition on MMEA
94.83Top-1 AccuracyCOMODO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| COMODOTrain=EgoExo4D, Supervised fine-tuning on source=false2025.03 | 94.83 | 99.47 | 99.7 | |
| COMODOTrain=Ego4D, Supervised fine-tuning on source=false2025.03 | 94.22 | 98.78 | 99.62 | |
| MantisSelf-supervised=false2025.03 | 93.01 | 98.25 | 99.01 | |
| IMU2CLIPTrain=Ego4D, Supervised fine-tuning on source=false2025.03 | 92.78 | 98.78 | 99.47 | |
| IMU2CLIPTrain=EgoExo4D, Supervised fine-tuning on source=false2025.03 | 92.78 | 98.93 | 99.7 | |
| COMODOSelf-supervised=true2025.03 | 92.48 | 99.01 | 99.77 | |
| MantisTrain=EgoExo4D, Supervised fine-tuning on source=true2025.03 | 92.25 | 98.1 | 99.16 | |
| MantisTrain=Ego4D, Supervised fine-tuning on source=true2025.03 | 91.72 | 98.48 | 99.32 | |
| IMU2CLIPSelf-supervised=true2025.03 | 91.41 | 98.71 | 99.47 | |
| MantisSelf-supervised=true2025.03 | 90.96 | 98.56 | 99.39 | |
| CrossHARSelf-supervised=false2025.03 | 84.97 | 96.55 | 97.74 | |
| CrossHARSelf-supervised=true2025.03 | 84.8 | 96.88 | 98.33 | |
| MOMENT-smallSelf-supervised=false2025.03 | 84.27 | 94.76 | 96.88 | |
| MOMENT-smallSelf-supervised=true2025.03 | 83.66 | 95.52 | 97.42 | |
| CrossHARTrain=EgoExo4D, Supervised fine-tuning on source=false2025.03 | 79.86 | 94.45 | 97.72 | |
| CrossHARTrain=Ego4D, Supervised fine-tuning on source=false2025.03 | 78.34 | 94.07 | 97.95 | |
| AttendSelf-supervised=false2025.03 | 58.66 | 79.87 | 87.47 | |
| DeepConvLSTMSelf-supervised=false2025.03 | 52.58 | 77.25 | 86.44 | |
| TimesNetSelf-supervised=false2025.03 | 50.3 | 74.24 | 83.99 | |
| InformerSelf-supervised=false2025.03 | 45.81 | 67.68 | 78.81 | |
| DLinearSelf-supervised=false2025.03 | 24.01 | 37.73 | 47.18 | |
| IMUGPT 2.0 (V+R)Self-supervised=false, Training Data Type=virtual + real IMU data2025.03 | 17.19 | 35.38 | 47.78 | |
| IMUGPT 2.0 (V)Self-supervised=false, Training Data Type=virtual-only IMU data2025.03 | 7.15 | 18.48 | 27.77 |