Virtual Category-Guided Continual Generalized Category Discovery

2026-07-06

AI Summary

This work addresses the challenges of unreliable pseudo-labels and model bias toward known classes in continual generalized category discovery, caused by ambiguous unlabeled samples. To mitigate these issues, the paper introduces Virtual Category Learning (VCL) into the continual learning setting for the first time, assigning uncertain samples to temporary virtual categories. This approach is further enhanced with Adaptive Boundary-Extended Neighborhood Contrastive Learning (ENCL), which collectively reduces prediction bias and improves the utilization of unlabeled data. The proposed method achieves substantial performance gains over existing approaches on CIFAR-100, Tiny ImageNet, and ImageNet-100, demonstrating more stable and scalable joint recognition capabilities for both known and novel categories.

Abstract

Continual Generalized Category Discovery (C-GCD) aims to incrementally identify novel categories from sequential unlabeled data while preserving recognition of known classes, which is an essential capability for open-world visual learning. A major bottleneck lies in ambiguous unlabeled samples that cannot be confidently assigned to known classes nor reliably grouped as novel ones, making pseudo-labeling brittle and often biasing learning toward familiar categories. In this work, we introduce Virtual Category-Guided Continual Generalized Category Discovery by adapting Virtual Category Learning (VCL) to the continual setting. Our method identifies uncertain samples and assigns them to temporary virtual categories, enabling safe and informative learning from unlabeled streams without injecting noisy labels, while improving unlabeled data utilization and mitigating prediction bias. To further stabilize discovery across sessions and enhance class separation, we augment VCL with Expanded Neighborhood Contrastive Learning (ENCL), which exploits extended neighborhood relations and an adaptive margin to learn more discriminative and well-separated representations for both old and emerging classes. Extensive experiments on CIFAR-100, Tiny ImageNet, and ImageNet-100 demonstrate that our approach consistently outperforms state-of-the-art methods, establishing a scalable and effective solution for C-GCD.

Problem

• Continual Generalized Category Discovery
• unlabeled data
• ambiguous samples
• pseudo-labeling
• open-world learning

Innovation

• Virtual Category Learning
• Continual Generalized Category Discovery
• Expanded Neighborhood Contrastive Learning
• Open-world Visual Learning
• Pseudo-labeling