A Unified Objective for Novel Class Discovery

TL;DR

This paper introduces a unified objective function for Novel Class Discovery that simplifies the learning process by combining supervised and unsupervised learning, leading to significant performance improvements on benchmark datasets.

Contribution

The paper proposes a novel unified objective (UNO) that replaces multiple loss functions with a single classification objective for discovering new classes.

Findings

UNO outperforms state-of-the-art methods by ~10% on CIFAR-100

UNO achieves +8% improvement on ImageNet

The approach simplifies NCD by integrating pseudo-labeling into a single framework

Abstract

In this paper, we study the problem of Novel Class Discovery (NCD). NCD aims at inferring novel object categories in an unlabeled set by leveraging from prior knowledge of a labeled set containing different, but related classes. Existing approaches tackle this problem by considering multiple objective functions, usually involving specialized loss terms for the labeled and the unlabeled samples respectively, and often requiring auxiliary regularization terms. In this paper, we depart from this traditional scheme and introduce a UNified Objective function (UNO) for discovering novel classes, with the explicit purpose of favoring synergy between supervised and unsupervised learning. Using a multi-view self-labeling strategy, we generate pseudo-labels that can be treated homogeneously with ground truth labels. This leads to a single classification objective operating on both known and unknown classes. Despite its simplicity, UNO outperforms the state of the art by a significant margin on several benchmarks (~+10% on CIFAR-100 and +8% on ImageNet). The project page is available at: https://ncd-uno.github.io.