Table 8 Effect of auxiliary task types and loss weights on material classification performance in multitask learning
From: Material classification method of traditional Chinese painting image based on prototypical network
Auxiliary task | w | Accuracy(%) | Precision(%) | Recall(%) | F1(%) |
|---|---|---|---|---|---|
Dynasty | 0.1 | 67.0 | 68.2 | 67.0 | 67.1 |
0.2 | 63.8 | 64.7 | 63.8 | 63.7 | |
0.3 | 69.2 | 70.8 | 69.2 | 69.2 | |
0.4 | 73.3 | 75.0 | 73.3 | 73.3 | |
0.5 | 79.8 | 79.8 | 79.8 | 79.8 | |
0.6 | 82.3 | 82.4 | 82.3 | 82.1 | |
0.7 | 82.4 | 82.7 | 82.4 | 82.3 | |
0.8 | 86.7 | 87.1 | 86.7 | 86.5 | |
0.9 | 80.9 | 83.0 | 80.9 | 80.9 | |
Content | 0.1 | 57.4 | 58.1 | 57.4 | 56.6 |
0.2 | 55.4 | 59.5 | 55.4 | 56.1 | |
0.3 | 63.0 | 63.7 | 63.0 | 61.7 | |
0.4 | 71.6 | 72.7 | 71.6 | 71.8 | |
0.5 | 79.8 | 80.1 | 79.8 | 79.6 | |
0.6 | 74.4 | 74.5 | 74.5 | 74.0 | |
0.7 | 76.8 | 76.9 | 76.8 | 76.5 | |
0.8 | 76.4 | 76.6 | 76.5 | 76.2 | |
0.9 | 78.9 | 79.0 | 78.9 | 78.6 | |
Technique | 0.1 | 35.8 | 34.1 | 35.8 | 33.2 |
0.2 | 63.3 | 65.9 | 63.3 | 63.9 | |
0.3 | 70.8 | 76.6 | 70.8 | 71.2 | |
0.4 | 70.5 | 74.2 | 70.5 | 71.5 | |
0.5 | 71.9 | 78.0 | 71.9 | 72.1 | |
0.6 | 74.2 | 76.8 | 74.2 | 74.8 | |
0.7 | 73.2 | 77.1 | 73.2 | 74.0 | |
0.8 | 66.7 | 72.7 | 66.7 | 68.1 | |
0.9 | 65.0 | 72.0 | 65.0 | 66.4 | |
None | 1.0 | 81.8 | 82.4 | 81.8 | 81.9 |
