training is slow !! - Githubissues

felixchenfy / Realtime-Action-Recognition

Apply ML to the skeletons from OpenPose; 9 actions; multiple people. (WARNING: I'm sorry that this is only good for course demo, not for real world applications !!! Those ary very difficult !!!)

MIT License

875 stars 256 forks source link

training is slow !! #79

Open Abd-Alrahman-Alahmar opened 2 years ago

Abd-Alrahman-Alahmar commented 2 years ago

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

felixchenfy commented 2 years ago

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

DukeZacharia commented 1 year ago

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