Bi-directional Feature Reconstruction Network for Fine-Grained Few-Shot Image Classification

The main challenge for fine-grained few-shot image classification is to learn feature representations with higher inter-class and lower intra-class variations, a mere few labelled samples. Conventional learning methods however cannot be naively adopted this setting -- quick pilot study reveals that they in fact push the opposite (i.e., variations variations). To alleviate problem, prior works predominately use support set reconstruct query then utilize metric determine its category. Upon careful inspection, we further reveal such unidirectional reconstruction only help increase are not effective tackling variations. In paper, first time introduce bi-reconstruction mechanism can simultaneously accommodate addition using increasing reducing This design effectively helps model explore more subtle discriminative features which key problem hand. Furthermore, also construct self-reconstruction module work alongside bi-directional make even discriminative. Experimental results on three widely used datasets consistently show considerable improvements compared other methods. Codes available at: https://github.com/PRIS-CV/Bi-FRN.