Human Activity Recognition on AnyMo Bench Fine150 Unseen Subject + Cross Device
24Top-1 AccuracyCOMODO
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| COMODOTraining epochs=20, IMU backbone=MantisV2, Video backbone=TimeSformer2026.05 | 24 | 50.6 | 8 | |
| MantisV2Training epochs=1002026.05 | 14.4 | 39.7 | 8.6 | |
| DeepConvLSTMTraining epochs=1002026.05 | 1.9 | 9.5 | 0.4 |