Novel Class Discovery (NCD) is a machine learning problem, where novel categories of instances are to be automatically discovered from an unlabelled pool. In contrast to clustering, NCD setting has access to labelled data from a disjoint set of classes. This topic has plausible real-world applications and is gathering much attention in the research community.
Here, we provide a non-exhaustive list of papers that studies NCD.
- None yet.
- Meta Discovery: Learning to Discover Novel Classes given Very Limited Data (ICLR 2022) [paper] [code]
- Self-Labeling Framework for Novel Category Discovery over Domains (AAAI 2022) [paper]
- Towards Open-Set Object Detection and Discovery (CVPR Workshop 2022) [paper]
- Divide and Conquer: Compositional Experts for Generalized Novel Class Discovery (CVPR 2022) [paper] [code]
- Novel Class Discovery in Semantic Segmentation (CVPR 2022) [paper] [code]
- Generalized Category Discovery (CVPR 2022) [paper] [code]
- Spacing Loss for Discovering Novel Categories (CVPR Workshop 2022) [paper]
- Novel Visual Category Discovery with Dual Ranking Statistics and Mutual Knowledge Distillation (NeurIPS 2021) [paper] [code] (DualRS)
- A Unified Objective for Novel Class Discovery (ICCV 2021) [paper] [code] (UNO)
- Joint Representation Learning and Novel Category Discovery on Single- and Multi-modal Data (ICCV 2021) [paper]
- Neighborhood Contrastive Learning for Novel Class Discovery (CVPR 2021) [paper] [code] (NCL)
- OpenMix: Reviving Known Knowledge for Discovering Novel Visual Categories in An Open World (CVPR 2021) [paper]
- Automatically Discovering and Learning New Visual Categories with Ranking Statistics (ICLR 2020) [paper] [TPAMI 2021] [code] (RS)
- Learning to discover novel visual categories via deep transfer clustering (ICCV 2019) [paper] [code] (DTC)
- Multi-class classification without multi-class labels (ICLR 2019) [paper] [code] (MCL)
Please help us improve the above listing by submitting PRs of other papers in this space. Thank you!