Multi-Content Interaction Network for Few-Shot Segmentation

TL;DR

This paper introduces MCINet, a novel network that enhances few-shot segmentation by exploiting multi-scale contextual interactions between support and query images, leading to state-of-the-art results especially on challenging datasets.

Contribution

The paper proposes a Multi-Content Interaction Network that fully exploits multi-scale features and bidirectional interactions to improve few-shot segmentation performance.

Findings

Achieves state-of-the-art results on benchmarks.

Outperforms competitors on the COCO dataset.

Effectively utilizes multi-scale contextual information.

Abstract

Few-Shot Segmentation (FSS) is challenging for limited support images and large intra-class appearance discrepancies. Most existing approaches focus on extracting high-level representations of the same layers for support-query correlations, neglecting the shift issue between different layers and scales, due to the huge difference between support and query samples. In this paper, we propose a Multi-Content Interaction Network (MCINet) to remedy this issue by fully exploiting and interacting with the multi-scale contextual information contained in the support-query pairs to supplement the same-layer correlations. Specifically, MCINet improves FSS from the perspectives of boosting the query representations by incorporating the low-level structural information from another query branch into the high-level semantic features, enhancing the support-query correlations by exploiting both the same-layer and adjacent-layer features, and refining the predicted results by a multi-scale mask prediction strategy, with which the different scale contents have bidirectionally interacted. Experiments on two benchmarks demonstrate that our approach reaches SOTA performances and outperforms the best competitors with many desirable advantages, especially on the challenging COCO dataset.