Giters

leftthomas / ESPCN

A PyTorch implementation of ESPCN based on CVPR 2016 paper "Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network"

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Discuss that the SISR recurring result is lower than the original paper?

FBLeee opened this issue · comments

commented

First of all ，thank you very much for your hard work to replicate the program！

The training set used in the original paper is 91 high-definition images, and the effect achieved by [×3_SR] is Set5—32.55dB.

The training set you used is 16700 images, and the reproduced effect is Set5—31.51dB. Why is the effect worse when using a larger training set?

And I found that the highest value of your program is Set5-34.09dB is higher than the original paper, so I guess the original paper uses the highest value as the result?

commented

@943301098 you could mail the author