However, a major drawback of this dual path architecture is the absence of communication between the two branches, leading to substantial semantic disparities in the encoded features. Merging high-level and low-level features directly could result in feature interference, consequently impeding adequate attention to objects of various scales during decoding and subsequently leading to performance deterioration.

Addressing Feature Interference in Dual Path Architectures for Improved Object Detection

原文地址: https://www.cveoy.top/t/topic/pz5W 著作权归作者所有。请勿转载和采集!

免费AI点我,无需注册和登录