ECCV 2026 Paper: Fourier Self-Supervision for Category Discovery

y_m_asano · x · 2026-08-31

Presents a paper accepted to ECCV 2026 titled "Fourier Self-Supervision for Fine-Grained Generalized Category Discovery" (FourEx).

Problem: Existing methods for Generalized Category Discovery, often based on self-supervision and contrastive learning, struggle to capture fine-grained distinctions, relying on superficial visual cues rather than intrinsic attributes.

Method:

Results: Experiments on multiple fine-grained datasets show that this approach outperforms state-of-the-art methods in discovering new categories and strengthening discriminative power, even when the number of classes is unknown.

Original post →

More from Research

Research channel →