Research Article
Multiscale Meets Spatial Awareness: An Efficient Attention Guidance Network for Human Parsing
Table 1
Embedding two types of Mutual Attention (MA) into four classical SPP models to evaluate them on the ATR test set.
| Method | Avg. F1 score | mIoU |
| DeepLab-v2 [10] | 73.21 | 58.50 | DeepLab-v2 + MA (parallel) | 73.36 | 61.01 | DeepLab-v2 + MA (sequential) | 73.94 | 61.43 | DeepLab-v3 [28] | 73.34 | 59.41 | DeepLab-v3 + MA (parallel) | 75.43 | 62.41 | DeepLab-v3 + MA (sequential) | 75.44 | 62.39 | PSPNet [11] | 73.46 | 59.27 | PSPNet + MA (parallel) | 75.27 | 62.15 | PSPNet + MA (sequential) | 75.41 | 62.70 | Vortex [12] | 72.96 | 58.55 | Vortex + MA (parallel) | 75.32 | 62.16 | Vortex + MA (sequential) | 75.49 | 62.97 |
|
|