DeepLabCut / DeepLabCut-live

SDK for running DeepLabCut on a live video stream
https://elifesciences.org/articles/61909
Other
192 stars 50 forks source link

Bump tensorflow version #110

Open stes opened 10 months ago

stes commented 10 months ago

See #103. Cannot foresee all possible side implications (windows eg), but this would be a useful change, as google colab now ships tf 2.15, and installing dlc live first re-installs TF on colab:

  Attempting uninstall: tensorflow
    Found existing installation: tensorflow 2.15.0
    Uninstalling tensorflow-2.15.0:
      Successfully uninstalled tensorflow-2.15.0

Would it maybe be possible to remove the restriction fully?

stes commented 10 months ago

Seems like this URL needs to be fixed first:

image

sneakers-the-rat commented 10 months ago

looks like previously there was reason to remove cap anyway :): https://github.com/DeepLabCut/DeepLabCut-live/pull/103#issuecomment-1584519710

MMathisLab commented 10 months ago

2.15 isn't compatible with dlc unfortunately... and now one needs to downgrade Cuda on Colab too (see open issues on main repo/PR from @n-poulsen. https://github.com/DeepLabCut/DeepLabCut/commit/258b19b427fdabf395e51266ea95d300e3ef5ad8

Plan is to roll out dlc3 this month to avoid all of this 🙌

AlexEMG commented 10 months ago

Yes, I think should open an issue clearly indicating at what TF/CUDA we stop here. And that new users should switch to: https://github.com/DeepLabCut/DeepLabCut-live-pytorch [WIP] 🚧

sneakers-the-rat commented 9 months ago

aha, i see last supported version of tf is 2.10: https://github.com/DeepLabCut/DeepLabCut/issues/2509#issuecomment-1933972939

Yes, I think should open an issue clearly indicating at what TF/CUDA we stop here. And that new users should switch to: https://github.com/DeepLabCut/DeepLabCut-live-pytorch [WIP] 🚧

is it time to set a deprecation warning and get ready to say farewell to this olde package then? i can prepare the horn section and champagne for the ceremony

MMathisLab commented 9 months ago

Yes! Actually we should set this warning when loaded; support ends at TF2.10 (for windows); and DLC is fully moving to pytorch 🤩🍾