search

speedinghzl / CCNet

CCNet: Criss-Cross Attention for Semantic Segmentation (TPAMI 2020 & ICCV 2019).
MIT License
1.42k stars 277 forks source link

The formula for criss-cross Attention #55

Closed Amorzhang closed 4 years ago

Amorzhang commented 4 years ago

image I recently read your paper and have some questions about the formula. What does the ∑ above mean? How does it work？ image image I found that the number of channels between them was not the same, while the number of channels to H‘ was C.

speedinghzl commented 4 years ago

A_{i,u} is a scalar. image

Amorzhang commented 4 years ago

Thanks