A Multi-head Attention Transformer with Convolutional Projections for Skin Cancer Image Classification
A hybrid approach that integrates convolutions with linear projections is introduced, combining the ability of convolutions to effectively capture local spatial features with the capacity of linear projections to model global relationships, leading to a more expressive and robust feature representation.