Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

v3 vs v4 speed comparison #7086

Closed
EmileTestUser opened this issue Dec 7, 2020 · 2 comments
Closed

v3 vs v4 speed comparison #7086

EmileTestUser opened this issue Dec 7, 2020 · 2 comments

Comments

@EmileTestUser
Copy link

Hi @AlexeyAB ,

I am currently writing my bachelor thesis in Munich about a comparison of version 3 and your well improved version 4. In order to fully understand the performance differences I have a question:

When comparing speed and accuracy of yolo v3 and v4 with my data I can notice a ~16% decrease of speed. As #6630 #5308 and Figure1 in the v4 paper prove, v4 speed will be lower for the same image size.

In contradiction, Table 8 in the paper shows the following values for a size of 416:
v4 - 38 FPS
v3 - 35 FPS

This would be an improvement of speed for the same image size. For other GPUs this is even higher.

What am I understanding wrong?

Thank you

@1993cathyzhao1993
Copy link

You mean the speed of v4 is faster than v3 based your data?

@AlexeyAB
Copy link
Owner

AlexeyAB commented Dec 7, 2020

  1. Speed in this repository was improved
    Table 8 in the paper shows the following values for a size of 416: https://arxiv.org/pdf/2004.10934.pdf
    v4 - 38 FPS - https://github.com/AlexeyAB/darknet
    v3 - 35 FPS - https://github.com/pjreddie/darknet

  2. If we use the same framework, then

  • YOLOv3 is slightly faster than YOLOv4 for the same resolution
  • YOLOv4 is many times faster than YOLOv3 for the same accuracy

https://arxiv.org/abs/2011.08036
scaled_yolov4_resolution - копия - копия

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants