Skip to content

Comments

Increase performance#55

Closed
ZxUkr wants to merge 3 commits intoserengil:masterfrom
ZxUkr:performance
Closed

Increase performance#55
ZxUkr wants to merge 3 commits intoserengil:masterfrom
ZxUkr:performance

Conversation

@ZxUkr
Copy link

@ZxUkr ZxUkr commented Jan 23, 2023

I adapted code for processing video. I added new function for batch processing. It also contains: filtering anchors before restore coordinates. And I cleared all unnecessary old reserved code. On my RTX 3070 I can processing 15 frame at once and it increase speed about 35%.

@serengil
Copy link
Owner

You should call detect_faces or extract_faces in a for loop to have this functionality. I am not sure about having dedicated function for this.

@serengil serengil closed this Dec 23, 2023
@emlcpfx
Copy link

emlcpfx commented Sep 4, 2024

Hey @ZxUkr, are you using RetinaFace on video? Did you build anything that incorporates temporal smoothing?

@DanielDimanov
Copy link

I don't get why this was closed if it works. Is there another way to do batch processing? Using a for loop vs paralelising with a batch is saving a good chunk of the processing time, which with videos is indeed still a concern as big GPUs are getting underutilised with RetinaFace.

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

Successfully merging this pull request may close these issues.

5 participants