Learning de-biased representations for remote-sensing imagery

• Main Authors: TIAN, Zichen, CHEN, Zhaozheng, SUN, Qianru
• Format: text
• Language: English
• Published: Institutional Knowledge at Singapore Management University 2024
• Subjects: SAR; Remote-sensing images; LoRA; Visual transformers; Categorical Data Analysis; Databases and Information Systems
• Online Access: https://ink.library.smu.edu.sg/sis_research/9400; https://ink.library.smu.edu.sg/context/sis_research/article/10400/viewcontent/2410.04546v1__1_.pdf
• Institution: Singapore Management University

Remote sensing (RS) imagery, requiring specialized satellites to collect and being difficult to annotate, suffers from data scarcity and class imbalance in certain spectrums. Due to data scarcity, training any large-scale RS models from scratch is unrealistic, and the alternative is to transfer pre-trained models by fine-tuning or a more data-efficient method LoRA. Due to class imbalance, transferred models exhibit strong bias, where features of the major class dominate over those of the minor class. In this paper, we propose debLoRA---a generic training approach that works with any LoRA variants to yield debiased features. It is an unsupervised learning approach that can diversify minor class features based on the shared attributes with major classes, where the attributes are obtained by a simple step of clustering.