Table 1 Accuracy of clip-level training and video-level training with different loss functions
From: Consistent constraint-based video-level learning for action recognition
Training method | Loss function | UCF101 (%) |
|---|---|---|
Clip level | Lossce | 51.52 |
Video level | Lossce | 53.44 |
| Â | \(\text {Loss}_{vll}\left (\text {Loss}^{err}_{ccl}\right)\) | 55.38 |
| Â | \(\text {Loss}_{vll}\left (\text {Loss}^{euc}_{ccl}\right)\) | 57.11 |
| Â | Lossvll(Lossccl) | 58.76 |