search

yun-liu / HAT-Net

Vision Transformers with Hierarchical Attention
96 stars 8 forks source link

About the upsampling method #1

Open mymuli opened 3 years ago

mymuli commented 3 years ago

May I ask, it is mentioned in formula (12) that the method of upsampling is used, which method is used specifically?

yun-liu commented 3 years ago

In implementation, the upsampling operation can be omitted if we just downsample the key tensor and the value tensor but keep the shape of the query tensor unchanged.