Secondly, inspired by the Adown subsampling convolution module from YOLOv9, we use it to replace the convolutions in the Backbone, enabling the model to capture finer image details at higher levels, ...