Research Article
Multilayer Convolutional Feature Aggregation Algorithm for Image Retrieval
Table 4
mAP values of different combinations of two layers in the fine-tuning model on the two datasets.
| ā | Pool4+Pool5 | Pool3+Pool4 | Pool3+Pool5 | Dimension | 1024 | 512 | 256 | 768 | 512 | 256 | 768 | 512 | 256 |
| Oxford5k+ML-CroW | 77.3 | 67.5 | 61.4 | 67.0 | 61.3 | 46.7 | 76.4 | 70.9 | 46.0 | Oxford5k+ML-RCroW | 79.5 | 78.0 | 74.1 | 69.9 | 68.4 | 63.3 | 78.7 | 78.1 | 74.8 | Paris6k+ML-CroW | 82.1 | 75.5 | 73.9 | 75.2 | 73.7 | 64.7 | 80.9 | 78.2 | 60.6 | Paris6k+ML-RCroW | 83.5 | 82.2 | 80.0 | 78.2 | 78.6 | 77.1 | 82.3 | 82.1 | 80.1 |
|
|