Advanced Configuration Options

faustomorales / keras-ocr

A packaged and flexible version of the CRAFT text detector and Keras CRNN recognition model.

https://keras-ocr.readthedocs.io/

MIT License

1.36k stars 347 forks source link

Closed neverabsolute closed 2 years ago

neverabsolute commented 2 years ago

can set MEMORY_GROWTH to any value to have Tensorflow use memory growth
can set MEMORY_ALLOCATED to any float to have Tensorflow restrict the memory usage as a ratio of the total available video memory

neverabsolute commented 2 years ago

Without memory growth and default weights:

With memory growth and default weights:

1.5GB VRAM taken up on a NVIDIA Tesla T4
running a concurrent Pytorch model (~2-3GB VRAM needed) works fine with plenty of headroom

neverabsolute commented 2 years ago

By default this doesn't actually affect existing users, this just adds options for more advanced users.

neverabsolute commented 2 years ago

That does work as expected, and I have implemented it. Thank you for the suggestion!

faustomorales commented 2 years ago

Thanks!

faustomorales commented 2 years ago

This new configuration code is now included in v0.9.0 of the package on PyPi. Thanks, again!

neverabsolute commented 2 years ago

Happy to help, this is huge for the work that I'm doing right now.