Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

MetaGCD: Learning to Continually Learn in Generalized Category Discovery

About

In this paper, we consider a real-world scenario where a model that is trained on pre-defined classes continually encounters unlabeled data that contains both known and novel classes. The goal is to continually discover novel classes while maintaining the performance in known classes. We name the setting Continual Generalized Category Discovery (C-GCD). Existing methods for novel class discovery cannot directly handle the C-GCD setting due to some unrealistic assumptions, such as the unlabeled data only containing novel classes. Furthermore, they fail to discover novel classes in a continual fashion. In this work, we lift all these assumptions and propose an approach, called MetaGCD, to learn how to incrementally discover with less forgetting. Our proposed method uses a meta-learning framework and leverages the offline labeled data to simulate the testing incremental learning process. A meta-objective is defined to revolve around two conflicting learning objectives to achieve novel class discovery without forgetting. Furthermore, a soft neighborhood-based contrastive network is proposed to discriminate uncorrelated images while attracting correlated images. We build strong baselines and conduct extensive experiments on three widely used benchmarks to demonstrate the superiority of our method.

Yanan Wu, Zhixiang Chi, Yang Wang, Songhe Feng• 2023

Related benchmarks

TaskDatasetResultRank
Generalized Category DiscoveryCUB
Accuracy (All)44.3
186
Category DiscoveryCUB-200 2011
Overall Score62.19
87
Category DiscoveryCIFAR-100
Accuracy (All Categories)55.49
39
Continual Category DiscoveryAverage fine-grained
cACC (All)65.38
32
Continual Category DiscoveryCaltech-101
cACC (All)83.05
16
Continual Category DiscoveryFGVC Aircraft
cACC (All)54.9
16
Continual Category DiscoveryStanford Cars
cACC (All)57.16
16
Continual Category DiscoveryTinyImageNet
cACC (All)56.15
16
Continual Category DiscoveryImageNet-100
cACC (All)70.2
16
Novel Class DiscoveryTiny-ImageNet
Accuracy (Seen)68.33
12
Showing 10 of 16 rows

Other info

Follow for update