LSKNet: A Foundation Lightweight Backbone for Remote Sensing

• DOI: 10.48550/arxiv.2403.11735
• Publication Date: 2024-03-18
• Authors: Yuxuan Li; Xiang Li; Yimain Dai; Qibin Hou; Li Liu; Yongxiang Liu; Mingming Cheng; Jian Yang
• Keywords: Foundation (evidence)

Remote sensing images pose distinct challenges for downstream tasks due to their inherent complexity. While a considerable amount of research has been dedicated remote classification, object detection and semantic segmentation, most these studies have overlooked the valuable prior knowledge embedded within scenarios. Such can be useful because objects may mistakenly recognized without referencing sufficiently long-range context, which vary different objects. This paper considers priors proposes lightweight Large Selective Kernel Network (LSKNet) backbone. LSKNet dynamically adjust its large spatial receptive field better model ranging context various in To our knowledge, selective kernel mechanisms not previously explored images. Without bells whistles, sets new state-of-the-art scores on standard segmentation benchmarks. Our comprehensive analysis further validated significance identified effectiveness LSKNet. The code is available at https://github.com/zcablii/LSKNet.