search

speedinghzl / CCNet

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

about the detail of cc-attention mathematical derivation #86

Closed pupil011 closed 3 years ago

pupil011 commented 3 years ago

i dont know why size of Ωu = (H+W-1)×C‘ how to got it (H+W-1) ? looking forward to your reply!

speedinghzl commented 3 years ago

@pupil011 For position u, the features of pixels in the same row or column with position u could form Ωu. There are W elements in the horizontal direction and H elements in the vertical direction. The position u is recorded twice. So the size of Ωu = (H+W-1)×C‘.

pupil011 commented 3 years ago

Thank you,i can clearly to understand . it's a wonderful  explanation!

---Original--- From: "Zilong Huang"<notifications@github.com> Date: Fri, Nov 20, 2020 22:37 PM To: "speedinghzl/CCNet"<CCNet@noreply.github.com>; Cc: "Mention"<mention@noreply.github.com>;"pupil011"<570756470@qq.com>; Subject: Re: [speedinghzl/CCNet] about the detail of cc-attention mathematical derivation (#86)

@pupil011 For position u, the features of pixels in the same row or column with position u could form Ωu. There are W elements in the horizontal direction and H elements in the vertical direction. The position u is recorded twice. So the size of Ωu = (H+W-1)×C‘.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.