Realtime-Action-Recognition icon indicating copy to clipboard operation
Realtime-Action-Recognition copied to clipboard

training is slow !!

Open Abd-Alrahman-Alahmar opened this issue 2 years ago • 2 comments

how do i increase the speed of the training ? it is too slow , also how to resume after pausing the training?

Abd-Alrahman-Alahmar avatar Mar 29 '22 11:03 Abd-Alrahman-Alahmar

  • I'm not sure... I think the training shouldn't be slow even using CPU, because the features used for training are human's positions, whose feature dimensionality is pretty small. It should just take seconds.
    • What is slow is detecting human skeletons from the training images. You'll need GPU to make it faster.
  • No, it's not supported --- I used the sklearn library's methods which don't support pausing/resuming the trainig

felixchenfy avatar May 14 '22 07:05 felixchenfy

I think the amount of the dataset you used may be the main issue. I used a large amount of dataset and it took 2 days to train

Reduce the amount of dataset and it may work in seconds

DukeZacharia avatar Mar 21 '23 01:03 DukeZacharia