search

argman / EAST

A tensorflow implementation of EAST text detector
GNU General Public License v3.0
3.02k stars 1.05k forks source link

About the size of data #73

Closed jiayi-wei closed 6 years ago

jiayi-wei commented 6 years ago

I am really confused about that the size of the score_map and geo_map are as same as the input image. However, pool5, which is 1/32 of the size of the input image, just is unsampled 3 times, which means that the size of final feature map is 1/4 of original input.

How could you get the loss between two map having different size?

jiayi-wei commented 6 years ago

Pardon me, I see that you take one every four in either map in icdar.py.