LogoNet: Layer-Aggregated Attention CenterNet for Logo Detection
Rahul Kumar Jain, Taro Watasue, Tomohiro Nakagawa, Takahiro Sato, Yutaro Iwamoto, Xiang Ruan, Yen‐Wei Chen · 2021
Logo detection is a key function for many applications. However, locating logos in complex scenarios is challenging and difficult due to the large variation of types and appearance of logos. In this paper, we propose a new deep learning logo detection algorithm called, LogoNet, the architecture of which includes hourglass like feature extraction backbone, a spatial attention module and a detection head same as CenterNet. The proposed LogoNet increases focus on logo objects more precisely within an image comparing with conventional algorithms. Our experiment results show approximately 1.5% improvement in performance compared to state-of-the-art anchor-free detection network on FlickrLogos-32 dataset.