Elimination of Non-Novel Segments at Multi-Scale for Few-Shot Segmentation

TL;DR

This paper introduces a multi-scale, scale-agnostic approach combined with an ensemble of base and meta learners to improve few-shot segmentation, effectively addressing spatial inconsistency and class bias issues.

Contribution

It proposes a novel method that simultaneously tackles spatial inconsistency and class bias in few-shot segmentation using multi-scale comparison and ensemble learning.

Findings

Achieves state-of-the-art results on PASCAL-5i dataset.

Outperforms previous methods on COCO-20i dataset.

Effectively reduces bias towards seen classes.

Abstract

Few-shot segmentation aims to devise a generalizing model that segments query images from unseen classes during training with the guidance of a few support images whose class tally with the class of the query. There exist two domain-specific problems mentioned in the previous works, namely spatial inconsistency and bias towards seen classes. Taking the former problem into account, our method compares the support feature map with the query feature map at multi scales to become scale-agnostic. As a solution to the latter problem, a supervised model, called as base learner, is trained on available classes to accurately identify pixels belonging to seen classes. Hence, subsequent meta learner has a chance to discard areas belonging to seen classes with the help of an ensemble learning model that coordinates meta learner with the base learner. We simultaneously address these two vital problems for the first time and achieve state-of-the-art performances on both PASCAL-5i and COCO-20i datasets.