LibFewShot: A Comprehensive Library for Few-Shot Learning

• DOI: 10.1109/tpami.2023.3312125
• Publication Date: 2023-09-05T17:47:12Z
• Authors: Wenbin Li; Ziyi Wang; Xuesong Yang; Chuanqi Dong; Pinzhuo Tian; Tiexin Qin; Jing Huo; Yinghuan Shi; Lei Wang; Yang Gao; Jiebo Luo
• Keywords: One shot; Contextual image classification

Few-shot learning, especially few-shot image classification, has received increasing attention and witnessed significant advances in recent years. Some studies implicitly show that many generic techniques or "tricks", such as data augmentation, pre-training, knowledge distillation, self-supervision, may greatly boost the performance of a learning method. Moreover, different works employ software platforms, backbone architectures input sizes, making fair comparisons difficult practitioners struggle with reproducibility. To address these situations, we propose comprehensive library for (LibFewShot) by re-implementing eighteen state-of-the-art methods unified framework same single codebase PyTorch. Furthermore, based on LibFewShot, provide evaluations multiple benchmarks various to evaluate common pitfalls effects training tricks. In addition, respect doubts necessity meta- episodic-training mechanism, our evaluation results confirm mechanism is still necessary when combined pre-training. We hope work can not only lower barriers beginners enter area but also elucidate nontrivial tricks facilitate intrinsic research learning.