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:
- Leverages the Fourier transform of images to enhance discrimination of subtle differences.
- Uses a dual frequency filtering strategy:
- Low-pass filter: Extracts broad, abstract attributes for high-level category info.
- High-pass filter: Emphasizes fine details like edges and textures for fine-grained recognition.
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.
More from Research
- Medusa from training to inference: a two-part guide to multi-token prediction acceleration — No_Progress_5399 · 2026-08-31
- LayerRecall: layer-wise memory routing for long-horizon consistency in video generation — zju · 2026-08-31
- EASEL benchmark: multimodal agents fail at dexterous, closed-loop visual tool use — EASEL-Bench · 2026-08-31
- StarHarness: evolving fixed-weight agent harnesses for enterprise tool use — ServiceNow-AI · 2026-08-31
- Paper: Deterministic Horizon Limits Pure Neural Reasoning — aronchick · 2026-08-31
- RAG poisoning causes 'attention collapse', fooling confidence detectors — rohanpaul_ai · 2026-08-31