We employ a self-supervised knowledge distillation model, DINOv2, as an image encoder to improve the feature representation of input images. To preserve the spatial relationships inherent within the ...