Relational Embedding for Few-Shot Classification

About

We propose to address the problem of few-shot classification by meta-learning "what to observe" and "where to attend" in a relational perspective. Our method leverages relational patterns within and between images via self-correlational representation (SCR) and cross-correlational attention (CCA). Within each image, the SCR module transforms a base feature map into a self-correlation tensor and learns to extract structural patterns from the tensor. Between the images, the CCA module computes cross-correlation between two image representations and learns to produce co-attention between them. Our Relational Embedding Network (RENet) combines the two relational modules to learn relational embedding in an end-to-end manner. In experimental evaluation, it achieves consistent improvements over state-of-the-art methods on four widely used few-shot classification benchmarks of miniImageNet, tieredImageNet, CUB-200-2011, and CIFAR-FS.

Dahyun Kang, Heeseung Kwon, Juhong Min, Minsu Cho · 2021

Related benchmarks

Task	Dataset	Result
Few-shot classification	tieredImageNet (test)	Accuracy—	282
5-way Few-shot Classification	MiniImagenet	Accuracy (5-shot)82.58	150
Few-shot Image Classification	miniImageNet (test)	Accuracy—	117
Few-shot classification	CUB	Accuracy—	104
5-way Few-shot Classification	CUB	Accuracy—	99
Few-shot classification	Mini-Imagenet 5-way 5-shot	Accuracy82.58	87
Few-shot classification	Mini-ImageNet 1-shot 5-way (test)	Accuracy67.6	82
5-way Classification	tieredImageNet (test)	Accuracy85.28	66
Few-shot classification	CUB-200 2011	5-way 1-shot Acc79.49	66
5-way Few-shot Classification	tieredImageNet	Accuracy (1-shot)71.61	49