A sine-guided Transformer–CNN network significantly outperformed 21 comparison methods on three benchmark datasets for ...
A DETR-CLIP study of 5,000 children’s drawings reported stronger detection and semantic performance than several baselines, ...