Inversed Pyramid Network with Spatial-adapted and Task-oriented Tuning for few-shot learning

Xiaowei Zhao, Duorui Wang, Shihao Bai, Shuo Wang, Yajun Gao, Yu Liang, Yuqing Ma, Xianglong Liu

Abstract: Highlights•TIPN merges global and local features in two stages for few-shot learning.•SA Layer preserves pretrained global features and learns adaptive local features.•Task-oriented Tuning bridges train-test class gap via task adaptation.•Extensive benchmarks validate superior performance over SOTA in few-shot learning.