You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Would it be possible to add TF 1.15 to the available versions in Fiji? This would enable support for CUDA 10.1.
New deep learning-based approaches are coming up, (YAPIC (on TF 2.0), StarDist in Qupath (on TF 1.15) ) that make use of TF > 1.14 are making it a bit of a nightmare to maintain different TensorFlow versions in a multi-user facility.
While switching TF versions is easy enough, switching the entire environment each time because they each depend on different CUDA versions is definitely a lot of work. There are workarounds of course, but this is not ideal...
I would be interested to hear your opinion and to know how easy it would be to add the TF 1.15 CPU and GPU libraries.
The text was updated successfully, but these errors were encountered:
Hi @lacan, the latest imagej-tensorflow-1.1.5 version already has the links to TF 1.15, can you give it a try? Because for me on Linux it did not work if I remember correctly, the 1.15 JNI was unpacking things differently than older TF versions and was not able to load, but that could be different for other OS. I'm sorry for being behind on updating the update sites, I am currently working on new releases and I expect to be done within the next two weeks.
I also wish handling the libraries was easier! I was wondering about programmatically checking the PATH.. If you have ideas, I'm happy for suggestions!
Hello,
Would it be possible to add TF 1.15 to the available versions in Fiji? This would enable support for CUDA 10.1.
New deep learning-based approaches are coming up, (YAPIC (on TF 2.0), StarDist in Qupath (on TF 1.15) ) that make use of TF > 1.14 are making it a bit of a nightmare to maintain different TensorFlow versions in a multi-user facility.
While switching TF versions is easy enough, switching the entire environment each time because they each depend on different CUDA versions is definitely a lot of work. There are workarounds of course, but this is not ideal...
I would be interested to hear your opinion and to know how easy it would be to add the TF 1.15 CPU and GPU libraries.
The text was updated successfully, but these errors were encountered: