Learning through Creation: A Hash-Free Framework for On-the-Fly Category Discovery

TL;DR

This paper introduces LTC, a hash-free, feature-based framework for on-the-fly category discovery that dynamically generates pseudo-unknown samples during training, improving unknown class detection and overall accuracy.

Contribution

LTC explicitly trains models for discovery tasks by jointly evolving a pseudo-unknown generator with the model, addressing optimization misalignment and enhancing representational capacity.

Findings

LTC outperforms prior methods across seven benchmarks.

Achieves 1.5% to 13.1% improvements in all-class accuracy.

Dynamic pseudo-unknown generation improves discovery performance.

Abstract

On-the-Fly Category Discovery (OCD) aims to recognize known classes while simultaneously discovering emerging novel categories during inference, using supervision only from known classes during offline training. Existing approaches rely either on fixed label supervision or on diffusion-based augmentations to enhance the backbone, yet none of them explicitly train the model to perform the discovery task required at test time. It is fundamentally unreasonable to expect a model optimized on limited labeled data to carry out a qualitatively different discovery objective during inference. This mismatch creates a clear optimization misalignment between the offline learning stage and the online discovery stage. In addition, prior methods often depend on hash-based encodings or severe feature compression, which further limits representational capacity. To address these issues, we propose Learning through Creation (LTC), a fully feature-based and hash-free framework that injects novel-category awareness directly into offline learning. At its core is a lightweight, online pseudo-unknown generator driven by kernel-energy minimization and entropy maximization (MKEE). Unlike previous methods that generate synthetic samples once before training, our generator evolves jointly with the model dynamics and synthesizes pseudo-novel instances on the fly at negligible cost. These samples are incorporated through a dual max-margin objective with adaptive thresholding, strengthening the model's ability to delineate and detect unknown regions through explicit creation. Extensive experiments across seven benchmarks show that LTC consistently outperforms prior work, achieving improvements ranging from 1.5 percent to 13.1 percent in all-class accuracy. The code is available at https://github.com/brandinzhang/LTC