Strengthening Spatial Relations to Multi-scale Features for Few-Shot Learning

研究成果: Conference contribution同行評審

摘要

Few-Shot Learning (FSL) is challenging due to limited training samples, hindering Convolutional Neural Networks (CNNs) from capturing discriminative object features. Recent approaches combine transfer learning and meta-learning, pre-training feature backbones on labeled base data and fine-tuning on novel data. In this work, we propose a CNN architecture with a cross-scale view of objects and introduce a spatial attention module to mitigate the impact of global average pooling. Our method achieves state-of-the-art results on standard benchmark for the cross-domain (CDFSL) few-shot task, demonstrating its effectiveness.

原文English
主出版物標題2023 5th International Conference on Computer Communication and the Internet, ICCCI 2023
發行者Institute of Electrical and Electronics Engineers Inc.
頁面56-59
頁數4
ISBN(電子)9798350326956
DOIs
出版狀態Published - 2023
事件5th International Conference on Computer Communication and the Internet, ICCCI 2023 - Fujisawa, 日本
持續時間: 23 6月 202325 6月 2023

出版系列

名字2023 5th International Conference on Computer Communication and the Internet, ICCCI 2023

Conference

Conference5th International Conference on Computer Communication and the Internet, ICCCI 2023
國家/地區日本
城市Fujisawa
期間23/06/2325/06/23

指紋

深入研究「Strengthening Spatial Relations to Multi-scale Features for Few-Shot Learning」主題。共同形成了獨特的指紋。

引用此