Questions about the composition of memory?

[hutao568]

Hi, thank you very much for your article. I have doubts about some of the details in the article. GKT uses the BEV(HWD) reference point to obtain the vision token from the image to form a VSCKK feature, then the Decoder's memory is HWD VSC Is the token composed of the characteristics of KK?

[hutao568]

您好，非常感谢您的文章。我对于文章中一些细节存在疑惑。 GKT利用BEV(HWD)参照点，去图像中获取vision token，组成一个VSCKK的特征，那么Decoder的memory是HWD 个VSCKK的特征组成的token吗