Huang Yinghua

Results 5 comments of Huang Yinghua

> I am very interested in this work,but i have a question about the selection of weight. How did you selecte the weight? It is the best performance in the...

> what happen to this, the 2d code keep request this for training Do you have a solution to this problem? Can you let me know?

> its a pretrained PTH file. just get it from the internet The link provided by the author in the 2D README section under "Model weights" regarding the "Learned weights"...

> 作者您好! 我注意到代码中的Attention模块里面存在一个论文中未标注的shortcut被用于残差连接。 ![Z{C1PKZCY)U{I0KNMD@SF@A](https://github.com/Visual-Attention-Network/VAN-Classification/assets/57658728/f52a6a2c-4d6a-4e0d-96bb-39a5677a4518) ![H)O7HCK4{8R2ZY2O$6@HUCD](https://github.com/Visual-Attention-Network/VAN-Classification/assets/57658728/df8a4300-a0e4-4cde-9c44-aaec3bc8f500) 想请问一下,这个连接是必要的吗 我在研究源码时一只无法解惑作者这个DW-Conv 和DW-D-Conv的实现是怎么实现的以及LKA怎么用到这两个东西,请问您知道? @GoPikachue

> 我理解的:DW-Conv用于通道计算,DW-D-Conv捕捉远距离依赖。[@mrbird643](https://github.com/mrbird643) 醍醐灌顶,非常感谢。