.png){kind=icon}
返回
Object detection based on few-shot learning via instance-level feature correlation and aggregation
.png){kind=icon}
PRE
AI
DOI：10.1007/s10489-022-03399-2.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
The detection of novel foregrounds only utilizing scarce annotated images, namely few-shot object detection, makes a detector no longer dependent on large-scale instantiated sets. The realistic challenge might lie in establishing the correlation of few instances and balancing the sensitivity between base and novel categories. In this paper, we propose a few-shot detector using instance-level feature correlation based on an interactive self-attention module to deeply mine the discriminating representations from scarce novel instances. Besides, using an extended soft threshold shrinkage, a feature aggregation procedure is introduced to eliminate redundant information while enhancing the representation sensitivity between base and novel categories. In the training phase, an orthogonal loss is applied to further enhance the feature distinguishability of inter-categories. Finally, we evaluate related competitive detectors on both benchmarks PASCAL-VOC07/12 and MS-COCO, with the results verifying the superior detection precision on AP, mAP and AR measurements of the proposed approach.
Keyword:
Few-shot learning
Object detection
Self-attention
Feature aggregation
Orthogonal loss
期刊
IF:
3.5
论文数:
7.6K
被引数:
1.7W
机构
暂无机构信息
引用论文
.png){kind=icon}
Do Perceptions of Competence Mediate The Relationship Between Fundamental Motor Skill Proficiency and Physical Activity Levels of Children in Kindergarten?能力的感知是否可以介导幼儿园儿童的基本运动技能熟练程度与身体活动水平之间的关系？
PREAI
分享收藏
查看更多内容