From patch, sample to domain: Capture geometric structures for few-shot learning

Published: 01 Jan 2024, Last Modified: 06 Feb 2025Pattern Recognit. 2024EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•A novel Hierarchical Optimal Transport Network with Attention (HOTA) for CD-FSL.•HOTA captures geometric structures at domain/feature/sample levels.•HOTA maintains discrimination and transferability during domain alignment.•Our method gets superior results in diverse benchmarks (supervised, unsupervised).
Loading