Incremental Few-Shot Semantic Segmentation via Embedding Adaptive-Update and Hyper-class Representation

TL;DR

This paper introduces EHNet, a novel approach for incremental few-shot semantic segmentation that leverages hyper-class and category embeddings to mitigate feature drift and overfitting, achieving state-of-the-art results.

Contribution

The paper proposes an embedding adaptive-update strategy and hyper-class representation to address catastrophic forgetting and overfitting in IFSS, advancing the field with a new model architecture.

Findings

EHNet outperforms existing methods on PASCAL-5i and COCO datasets.

The hyper-class embedding effectively alleviates overfitting with limited samples.

Adaptive category embedding updates maintain knowledge across sessions.

Abstract

Incremental few-shot semantic segmentation (IFSS) targets at incrementally expanding model's capacity to segment new class of images supervised by only a few samples. However, features learned on old classes could significantly drift, causing catastrophic forgetting. Moreover, few samples for pixel-level segmentation on new classes lead to notorious overfitting issues in each learning session. In this paper, we explicitly represent class-based knowledge for semantic segmentation as a category embedding and a hyper-class embedding, where the former describes exclusive semantical properties, and the latter expresses hyper-class knowledge as class-shared semantic properties. Aiming to solve IFSS problems, we present EHNet, i.e., Embedding adaptive-update and Hyper-class representation Network from two aspects. First, we propose an embedding adaptive-update strategy to avoid feature drift, which maintains old knowledge by hyper-class representation, and adaptively update category embeddings with a class-attention scheme to involve new classes learned in individual sessions. Second, to resist overfitting issues caused by few training samples, a hyper-class embedding is learned by clustering all category embeddings for initialization and aligned with category embedding of the new class for enhancement, where learned knowledge assists to learn new knowledge, thus alleviating performance dependence on training data scale. Significantly, these two designs provide representation capability for classes with sufficient semantics and limited biases, enabling to perform segmentation tasks requiring high semantic dependence. Experiments on PASCAL-5i and COCO datasets show that EHNet achieves new state-of-the-art performance with remarkable advantages.