期刊:2013 Fourth International Conference on Computing, Communications and Networking Technologies (ICCCNT) [Institute of Electrical and Electronics Engineers] 日期:2020-07-01卷期号:: 1-6被引量:1
标识
DOI:10.1109/icccnt49239.2020.9225379
摘要
In this paper, we look at different architectures of feature pyramid networks and introduce a new one, namely, QuadFPN. We apply this FPN technique to the popular EfficientDet and obtain an improvement in performance by a minimum of 1.2% mAP on COCO dataset on each of the models under 10 million parameters (D0, D1, D2). Thus, we present a scalable and an alternate technique similar to BiFPN.