Organizing Background to Explore Latent Classes for Incremental Few-shot Semantic Segmentation

TL;DR

This paper introduces OINet, a novel approach for incremental few-shot semantic segmentation that organizes background embeddings and inherits prototypes to effectively learn new classes with limited data while preventing forgetting.

Contribution

The paper proposes a new network architecture, OINet, which organizes background embeddings and inherits prototypes to improve incremental learning in few-shot segmentation.

Findings

Achieves state-of-the-art results on Pascal-VOC and COCO datasets.

Effectively prevents catastrophic forgetting during incremental learning.

Successfully learns new classes with few samples without disrupting base class representations.

Abstract

The goal of incremental Few-shot Semantic Segmentation (iFSS) is to extend pre-trained segmentation models to new classes via few annotated images without access to old training data. During incrementally learning novel classes, the data distribution of old classes will be destroyed, leading to catastrophic forgetting. Meanwhile, the novel classes have only few samples, making models impossible to learn the satisfying representations of novel classes. For the iFSS problem, we propose a network called OINet, i.e., the background embedding space \textbf{O}rganization and prototype \textbf{I}nherit Network. Specifically, when training base classes, OINet uses multiple classification heads for the background and sets multiple sub-class prototypes to reserve embedding space for the latent novel classes. During incrementally learning novel classes, we propose a strategy to select the sub-class prototypes that best match the current learning novel classes and make the novel classes inherit the selected prototypes' embedding space. This operation allows the novel classes to be registered in the embedding space using few samples without affecting the distribution of the base classes. Results on Pascal-VOC and COCO show that OINet achieves a new state of the art.