Hybrid AI With Attention Achieves Record Face Expression Recognition Accuracy
A Moroccan research team has developed a hybrid convolutional neural network with local and global attention modules that reaches 95.87 ...
A Moroccan research team has developed a hybrid convolutional neural network with local and global attention modules that reaches 95.87 ...
Researchers have developed an object-state-based frame-sampling method that trains video object detectors up to 7.7 times faster than full-data training ...
Researchers have developed a unified scoring method that combines multiple accuracy metrics with model confidence to provide a more clinically ...
Researchers at the Norwegian University of Science and Technology have developed two new explainability methods that integrate transformer attention weights ...
Researchers have developed CNN-SA-RFR, a cosine similarity-based self-attention framework that suppresses redundant filters during CNN training, boosting plant disease classification ...
A new robust learning framework from Zhejiang University addresses identity misalignment in unsupervised visible-infrared person re-identification, improving cross-modality retrieval under ...
Researchers have developed a single-arm robotic sewing system that combines computer vision, neural-network grasp estimation, and force control to handle ...
A new explainable deep learning framework segments marine biofouling at the pixel level, improving underwater inspection accuracy and structural health ...
A new special issue in Medical & Biological Engineering & Computing showcases how deep learning is transforming ultrasound image segmentation, ...
A new Communications Engineering study formalizes when frame-based and event-based visual data each hold a fundamental information advantage and how ...
© 2026 Scienmag - Science Magazine
© 2026 Scienmag - Science Magazine