K.B. Sivangi and F. Deligianni BMVC 2024 pdf, Project Page and Source Code
By distilling knowledge from the teacher model, the student network learns to predict keypoints with high accuracy while maintaining lower computational complexity. This approach significantly advances the field by balancing efficiency and accuracy. We specifically employ Global Filter Layers, which operate in the frequency domain, to reduce the processing overhead associated with traditional attention mechanisms. Our method was evaluated using both static and dynamic filter weighting strategies, demonstrating that the Global Filter Layers not only improve speed but also maintain a competitive level of accuracy compared to traditional attention-based models. …read more
I. Zakariyya, L. Tran, K.B. Sivangi, F. Deligianni, submitted
Privacy Preservation in Radar-based Human Activity Recognition
Q. Liu, P. Henderson, X. Gu, H. Dai, submitted
We are the first to propose a registration-guided method for semi-supervised medical image segmentation, by integrating registration with a contrastive cross-teaching framework. Furthermore, we introduce a novel registration supervision loss that enhances cross-teaching, by providing additional and informative registered pseudo-labels early in training, automatically selecting the best registered volumes.more
Q. Liu, X. Gu, P. Henderson, BMVC 2023. BMVC2023_arxiv, Project Page and Source Code, Extended Version
We develop a novel Multi-Scale Cross Supervised Contrastive Learning (MCSC) framework, to segment structures in medical images. We jointly train CNN and Transformer models, regularising their features to be semantically consistent across different scales. To tackle class imbalance, we take into account the prevalence of each class to guide contrastive learning and ensure that features adequately capture infrequent classes.
Q. Liu, C. Kaul, J. Wang, C. Anagnostopoulos, R. Murray-Smith, F. Deligianni, IEEE International Conference on Acoustics, Speech, and Signal Processing, 2023. paper
We design of a compact and accurate Transformer network for Medical Image Semantic Segmentation, which introduces convolutions in a multi-stage design for hierarchically enhancing spatial and local modeling ability of Transformers. This is mainly achieved by our well-designed Convolutional Swin Transformer (CST) block which merges convolutions with Multi-Head Self-Attention and Feed-Forward Networks for providing inherent localized spatial context and inductive biases.
D. Szczepaniak, M. Harvey, F. Deligianni, IEEE International Symposium on Mixed and Augmented Reality, 2024. paper, Project Page
We employed machine learning techniques to enable real-time, personalized cognitive training. This work contributes to the development of more effective cognitive training interventions that can adapt to individual differences and maintain optimal engagement levels.
N. Lai-Tan, M. Philiastides, F. Deligianni, International Conference on Affective Computing and Intelligent Interaction (ACII 2024), Best Student Paper Runner-up Award, 2024. paper
Fusion of Spatial and Riemannian Features to Enhance Detection of Gait Adaptation Mental States During Rhythmic Auditory Stimulation
M. Alkan, G. Veldtman, F. Deligianni, IEEE International Symposium on Biomedical Imaging (IEEE ISBI), 2024 arxiv, paper, code
Here, we exploit the Riemannian geometry of the spatial covariance structure of the ECG signal to improve classification in extremely heterogeneous and small datasets.