SETR-pytorch
SETR-pytorch copied to clipboard
doubt about patch embeddings
Hi~awsome repo. But I wonder if it is necessary to implement GELU and LayerNorm after linear layer to get patch embedding. Neither the ViT paper and code applies these layers. What I mentioned is in Line 269 and 270 in /SETR/transformer_model.py
transformer 文章里面是使用了这两个操作,你可以去看下原文中的模型结构。 Attention is all you need.
但ViT的文章和代码好像没有用
这个我也不清楚了 说实话我没看过那篇文章跟代码... 你可以去掉试试 我觉得影响不大应该。
好的,多谢