首页swin_transformer

swin_transformer

时间: 2023-10-28 15:43:42 浏览: 75

Swin-Transformer

Swin-Transformer是2021年微软研究院发表在ICCV上的一篇文章，并且已经获得ICCV 2021 best paper的荣誉称号。虽然Vision Transformer (ViT)在图像分类方面的结果令人鼓舞，但是由于其低分辨率特性映射和复杂度随图像大小的二次增长，其结构不适合作为密集视觉任务或高分辨率输入图像的通过骨干网路。为了最佳的精度和速度的权衡，提出了Swin-Transformer结构。

The SWin Transformer is a type of transformer architecture that was introduced in 2021 by Microsoft Research Asia. It is a variant of the standard transformer architecture that is designed to better handle images and other large inputs. The SWin Transformer achieves this by introducing a new type of attention mechanism that is optimized for images, as well as a new type of block called a Switchable Kernel block that allows the model to adjust its receptive field size based on the input. The SWin Transformer has achieved state-of-the-art results on several image classification tasks, including the popular ImageNet dataset, with fewer parameters and faster training times than previous state-of-the-art models. It has also shown promising results on other computer vision tasks such as object detection and semantic segmentation. Overall, the SWin Transformer represents an exciting new direction in transformer-based models for computer vision, and has the potential to significantly improve the performance of a wide range of image-based tasks.

阅读全文