Transductive Meta-Learning with Enhanced Feature Ensemble for Few-shot Semantic Segmentation

Abstract

This paper addresses few-shot semantic segmentation and proposes a novel transductive end-to-end method that overcomes three key problems affecting performance. First, we present a novel ensemble of visual features learned from pretrained classification and semantic segmentation networks with the same architecture. Our approach leverages the varying dis-criminative power of these networks, resulting in visual features that capture rich and diverse information at different depths. Secondly, the pretrained semantic segmentation network serves as a base class extractor, which significantly reduces false positives. Thirdly, a transductive meta-learning method addresses the prediction by first learning the relationship between labeled and unlabeled data points with matching support foreground to query features (intra-class similarity) and then applying this knowledge to predict on the unlabeled query image (intra-object similarity), which simultaneously learns propagation and false positive suppression. To evaluate our method, we performed experiments on benchmark datasets, and the results demonstrate significant improvement, with minimal trainable parameters 2.98M. Specifically, using Resnet-101, we achieve state-of-the-art performance for both 1-shot and 5-shot Pascal-5i , as well as on 1-shot and 5-shot COCO-20i .