Dynamic Convolution and Graph-Coupled Attention for Cross-Subject EEG-Vision Decoding


Tianyu Zhang (Durham University), FAN WAN (Tongfang Knowledge Network Digital Technology Co., Ltd. China National Nuclear Corporation), Kaili Sun (Durham University), Xingyu Miao (Durham University), Yueming Sun (University of Durham), Minye Shao (Durham University), Yang Long (Durham University)
The 35th British Machine Vision Conference
PDFPosterVideo (Right click to download)

Abstract

Electroencephalography (EEG) offers a non-invasive route to visual object decoding, but practical deployment is hampered by the signals’ non-stationarity, low signal-to-noise ratio and pronounced inter-subject variability. Existing models employ fixed convolutional filters and therefore generalize poorly across subjects. We introduce \textit{ECHO-Net}—an adaptive, hierarchically organised network that assembles dynamic convolutional kernels, conditioning them on each incoming EEG trial to capture transient neural dynamics. A cross-modal contrastive objective aligns the resulting representations with CLIP image embeddings, while a channel–filter attention mechanism emphasises task-relevant electrodes and time–frequency bands. To regularise spatial structure, an embedded EEG-GAT module propagates information over a fully connected electrode graph, producing more consistent cross-subject features. Evaluated on the 200-way THINGS-EEG benchmark, our method attains 18.5\% top-1 and 44.1\% top-5 cross-subject accuracy—surpassing the strongest prior approach by 2.9\% top-1.

Citation

@inproceedings{Zhang_2025_BMVC,
author    = {Tianyu Zhang and FAN WAN and Kaili Sun and Xingyu Miao and Yueming Sun and Minye Shao and Yang Long},
title     = {Dynamic Convolution and Graph-Coupled Attention for Cross-Subject EEG-Vision Decoding},
booktitle = {36th British Machine Vision Conference 2025, {BMVC} 2025, Sheffield, UK, November 24-27, 2025},
publisher = {BMVA},
year      = {2025},
url       = {https://bmva-archive.org.uk/bmvc/2025/assets/papers/Paper_509/paper.pdf}
}


Copyright © 2025 The British Machine Vision Association and Society for Pattern Recognition
The British Machine Vision Conference is organised by The British Machine Vision Association and Society for Pattern Recognition. The Association is a Company limited by guarantee, No.2543446, and a non-profit-making body, registered in England and Wales as Charity No.1002307 (Registered Office: Dept. of Computer Science, Durham University, South Road, Durham, DH1 3LE, UK).

Imprint | Data Protection