search

arsfutura / face-recognition

A framework for creating and using a Face Recognition system.
BSD 3-Clause "New" or "Revised" License
146 stars 49 forks source link

using gpu? #25

Closed crypt0miester closed 3 years ago

crypt0miester commented 3 years ago

hey, do you have a version of this using gpu?

ldulcic commented 3 years ago

Hi @crypt0miester! There is no support for the GPU at the moment, I do plan to add it in the future, it shouldn't be too much work. PRs are welcome if you are willing to help me 😊

crypt0miester commented 3 years ago

I'll try my best. currently setting up my multiple amd rx 570. takes time and energy to build torch with it. lol

crypt0miester commented 3 years ago

for anyone reading this in the future. the code works amazingly in CPU. no need for a GPU.

the time taken using a GPU for a small number of classes are higher than using CPU. you may argue that my GPU (RX 570) is old, but the reason the code does not need GPU is that you can retrain the model with older dataset. makes it much faster and convenient.

thanks @ldulcic.

ldulcic commented 3 years ago

No problem @crypt0miester, glad you made it work 😊 Is the library working good for you?

crypt0miester commented 3 years ago

yes, very well. Thank you very much.

I have put a necessity of 30 images per class.

Currently working on filtering "bad" images so that I get a good accuracy for future inference. So if bad probability of an image detected it would be removed and new images would be added.

Been scratching my brain for hours.

Do you have an idea of how to implement that?

ldulcic commented 3 years ago

Currently working on filtering "bad" images so that I get a good accuracy for future inference. So if bad probability of an image detected it would be removed and new images would be added.

@crypt0miester how do you decide which new images would you add?