EDSR-PyTorch icon indicating copy to clipboard operation
EDSR-PyTorch copied to clipboard

Published 20 hours ago • verified publisher icon sanghyun-son

Running speed.

Open xiaozhi2015 opened this issue 5 years ago • 3 comments

Can you share your running speed when inferring, for example, upscale a 1920 * 1080 img to 3840 * 2160? @thstkdgus35 @yulunzhang

xiaozhi2015 avatar Jun 25 '19 07:06 xiaozhi2015

Hello @xiaozhi2015,

Actually, it is not straightforward to measure the running speed, because it depends on

  • CUDA initialization time

  • GPU API call overheads

  • Overheads from reading/writing large images from a disk

Therefore, the running time can vary depending on your situation.

In general, if you try the code over >100 images and average the running time, you can find an approximate running speed.

I will also try that and report to you if possible.

Thank you!

sanghyun-son avatar Jul 01 '19 01:07 sanghyun-son

Much thx for your reply @thstkdgus35. What I want to know is the pure inferring time(exclude model initialization, read img, save img). Looking forward to your report.

xiaozhi2015 avatar Jul 01 '19 03:07 xiaozhi2015

Hello @xiaozhi2015,

Actually, it is not straightforward to measure the running speed, because it depends on

  • CUDA initialization time
  • GPU API call overheads
  • Overheads from reading/writing large images from a disk

Therefore, the running time can vary depending on your situation.

In general, if you try the code over >100 images and average the running time, you can find an approximate running speed.

I will also try that and report to you if possible.

Thank you! hi @sanghyun-son i facing the problem at https://github.com/sanghyun-son/EDSR-PyTorch/issues/319#issue-996722398 could you help me?

ulfailliyina avatar Sep 15 '21 07:09 ulfailliyina