Abstract | ||
---|---|---|
Scene recognition aims at classifying a scene image to one of the predefined scene categories by comprehending the entire image. The complex composition of scenery images makes scene recognition a challenging task. However, most state-of-the-art visual recognition methods are developed on general-purpose datasets and omit the uniqueness of scene data. In this work, we propose an efficient Scale Attentive (SA) Module to address the predicament of scene recognition, which streamlines the scale-aware attention learning pipeline to assist the feature re-calibration and refinement process. By integrating SA Module into ResNet-50, we obtain a boost of Top-1 accuracy by 1.83% on the benchmark scene dataset with only 0.12% additional parameters and 0.24% additional FLOPs. Moreover, comprehensive experiments show that our method achieves better performance compared with the state-of-the-art attention and multi-scale methods in a computationally efficient manner. |
Year | DOI | Venue |
---|---|---|
2022 | 10.1016/j.neucom.2021.12.053 | Neurocomputing |
Keywords | DocType | Volume |
Convolutional Neural Network,Scene recognition,Attention,Multi-scale learning | Journal | 492 |
ISSN | Citations | PageRank |
0925-2312 | 0 | 0.34 |
References | Authors | |
0 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Xiao-Hui Yuan | 1 | 534 | 75.44 |
Zhinan Qiao | 2 | 0 | 2.03 |
Abolfazl Meyarian | 3 | 0 | 0.34 |