Table 8 Sensitivity analysis for contextual stream parameters on PoseTrack dataset.
From: Learning spatio-temporal context for basketball action pose estimation with a multi-stream network
Model type | Sequence length | Hidden units | Hip AP (%) | Total AP (%) | MAP (%) |
|---|---|---|---|---|---|
LSTM | 30 frames | 128 | 75.4 | 74.8 | 75.6 |
LSTM | 60 frames | 128 | 76.2 | 75.5 | 76.3 |
LSTM | 60 frames | 256 | 76.8 | 76.1 | 76.9 |
Transformer | 30 frames | 128 | 76.0 | 75.2 | 76.0 |
Transformer | 60 frames | 128 | 76.7 | 76.4 | 77.1 |
Transformer | 60 frames | 256 | 77.3 | 77.0 | 77.8 |
