arxiv_cv 90% Match Research Paper Researchers in Computer Vision,Machine Learning Engineers,Data Scientists,AI Researchers working on unsupervised/semi-supervised learning 2 weeks ago

SEAL: Semantic-Aware Hierarchical Learning for Generalized Category Discovery

computer-vision › scene-understanding

📄 Abstract

Abstract: This paper investigates the problem of Generalized Category Discovery (GCD). Given a partially labelled dataset, GCD aims to categorize all unlabelled images, regardless of whether they belong to known or unknown classes. Existing approaches typically depend on either single-level semantics or manually designed abstract hierarchies, which limit their generalizability and scalability. To address these limitations, we introduce a SEmantic-aware hierArchical Learning framework (SEAL), guided by naturally occurring and easily accessible hierarchical structures. Within SEAL, we propose a Hierarchical Semantic-Guided Soft Contrastive Learning approach that exploits hierarchical similarity to generate informative soft negatives, addressing the limitations of conventional contrastive losses that treat all negatives equally. Furthermore, a Cross-Granularity Consistency (CGC) module is designed to align the predictions from different levels of granularity. SEAL consistently achieves state-of-the-art performance on fine-grained benchmarks, including the SSB benchmark, Oxford-Pet, and the Herbarium19 dataset, and further demonstrates generalization on coarse-grained datasets. Project page: https://visual-ai.github.io/seal/

Authors (3)

Zhenqi He

Yuanpei Liu

Kai Han

Submitted

October 21, 2025

arXiv Category

cs.CV

arXiv PDF

Key Contributions

Introduces SEAL, a Semantic-aware Hierarchical Learning framework for Generalized Category Discovery (GCD). SEAL leverages naturally occurring hierarchical structures and employs Hierarchical Semantic-Guided Soft Contrastive Learning and a Cross-Granularity Consistency module to effectively categorize both known and unknown classes from partially labeled data.

Business Value

Enables more robust and scalable image categorization systems that can adapt to new, unseen categories, which is valuable for large-scale image databases, content filtering, and autonomous systems that need to identify novel objects or situations.

Paper Metadata

Innovation Type

Algorithmic Improvement and Framework Design

Deployment Feasibility

Moderate. Requires careful implementation of hierarchical structures and contrastive learning, but the framework aims for generalizability.

Limitations Addressed

Limited generalizability and scalability of existing GCD approaches relying on single-level semantics or manual hierarchies,Ineffectiveness of conventional contrastive losses in handling hierarchical relationships,Difficulty in aligning predictions across different levels of granularity

Technical Tags

Generalized Category Discovery (GCD)semantic-aware learninghierarchical learningcontrastive learningsoft negativescross-granularity consistencypartially labelled dataunknown classesrepresentation learninghierarchical structure

Research Topics

Unsupervised LearningSemi-Supervised LearningCategory DiscoveryRepresentation LearningHierarchical Classification

Methods & Architectures

SEAL frameworkHierarchical Semantic-Guided Soft Contrastive LearningCross-Granularity Consistency (CGC) moduleExploiting hierarchical similarityGenerating soft negatives Hierarchical Learning ModelsContrastive Learning Models

Applications & Tasks

Image Recognition Data Mining Content Moderation Robotics Perception Autonomous Systems Category DiscoveryClassificationRepresentation LearningHandling Unknown Classes Generalized Category Discovery (GCD)Categorizing images with unknown classes

Related Fields

Computer VisionMachine LearningUnsupervised LearningData MiningOntology Engineering

Keywords

Generalized Category DiscoveryGCDsemantic-awarehierarchical learningcontrastive learningunknown classespartially labeledrepresentation learningSEALcomputer visionunsupervised learning

Academic Context

#Unsupervised Learning#Semi-Supervised Learning#Category Discovery#Representation Learning#Hierarchical Classification

Commercial Potential

Potential Products

Scalable image tagging and categorization systemsContent moderation tools that can identify new types of harmful contentAutonomous systems capable of recognizing novel objects

Target Industries

Social MediaE-commerceContent ManagementRoboticsSecurity

Use Case Examples

Automatically categorizing a large, unlabeled image datasetDeveloping a system that can identify new types of objects in real-timeBuilding a content moderation system that adapts to emerging trends

Competitive Edge

Offers a more generalizable and scalable approach to GCD by incorporating semantic awareness and hierarchical structures, overcoming limitations of previous methods.

Market Opportunity

Growing need for automated categorization and understanding of large visual datasets.

Revenue Models

Licensing of the technologyintegration into AI platforms for data analysis.

Resource Requirements

Compute Needs

Significant, for training deep learning models with contrastive learning and hierarchical structures.

Data Requirements

Partially labeled datasets, ideally with some inherent hierarchical structure.

Deployment Constraints

Requires careful design of the hierarchical structure and effective training strategies.

Scalability

The framework aims for improved scalability by leveraging hierarchical structures and efficient learning techniques.

Production Readiness

Maturity Level

Research/Development

Time to Market

2-4 years for integration into robust systems.

Patent Potential

Moderate, the SEAL framework and its specific learning objectives could be patentable.

View Full Paper Back to Papers