Feature Distribution Fitting with Direction-Driven Weighting for Few-Shot Images Classification

Few-shot learning has received increasing attention and witnessed significant advances in recent years. However, most of the few-shot methods focus on optimization training process, metric sample generating networks. They ignore importance ground-truth feature distributions classes. This paper proposes a direction-driven weighting method to make classes precisely fit distributions. The learned can generate an unlimited number samples for avoid overfitting. Specifically, proposed consists two strategies. strategy is capturing more complete direction information that describe similarity-weighting estimate impact different fitting procedure assign corresponding weights. Our outperforms current state-of-the-art performance by average 3% 1-shot standard benchmarks like miniImageNet, CIFAR-FS, CUB. excellent compelling visualization show our accurately