Giters

proteus1991 / GridDehazeNet

This repo contains the official training and testing codes for our paper: GridDehazeNet: Attention-Based Multi-Scale Network for Image Dehazing.

Home Page:https://jhc.sjtu.edu.cn/~xiaohongliu/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

some issues about code

liang233 opened this issue · comments

commented

hello, I have some issues about code,
image

1.how do the parameter “output_size” work when the size of feature map is odd? by zero padding or other ways？
2. If zero padding is not added during training and zero padding needs to be added during testing, will there be performance degradation?

thanks

commented

Hi, I will try to answer your questions as follows.

  1. "output_size" is a parameter to give the exact output size you defined. To my understanding, it is similar to the "same" parameter in TensorFlow, which is achieved by zero padding as default.
  2. I didn't test this case in detail. Intuitively, I believe there exists performance degradation, but it may be negligible.

If you have any other question, please let me know.