New TagLab version update issues

[wogreene]

Hi all -

Great to see a new update come out! In attempting to upgrade, I've run into issues. Because it now requires Python 3.11, I've had to rebuild all dependencies (we were using python 3.9 previously). In the install script, it's seeming like it is mandating torch 1.13.1, but there are no wheels out there for that version of torch and python 3.11. For reference, I'm running CUDA 11.6 (what we found worked best with the previous release of TagLab), but I'm not sure that's the issue here. If you all could provide some guidance as to what we should do about torch version / python version incompatibility, along with which version of CUDA is preferred for the new release, that would be great. Thanks!

-Will

[Jordan-Pierce]

I believe the updated install.py script has not yet been pushed from the devel branch to main.

[wogreene]

Hey Jordan, sorry for jumping the gun on this! We use taglab all the time within our org and I got like 10 slack messages from coworkers the minute you pushed the update notifying me that they couldn’t get it to work. Can I just use the install script from devel or do I need to wait on another push? Will

…

On Tue, Oct 29, 2024 at 3:48 PM Jordan Pierce ***@***.***> wrote: I believe the updated install.py script has not yet been pushed from the devel branch to main. — Reply to this email directly, view it on GitHub <#161 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/A35RBYIN3RGG6GYE4OOXKMDZ57RBHAVCNFSM6AAAAABQ2TU5YGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDINBVGE4TCOJXGE> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[maxcorsini]

Hi all, it seems that for some reason the install.py is the old one. Now it should work. Run update.py and then install.py. Pay attention that now TagLab supports Python 3.11 only.

Max

[Jordan-Pierce]

@wogreene ensure that when installing w/ 3.11 you specify the SAM flag:

python install.py SAM

As it's an optional addition, but currently needs it.

@massimomartinelli might I recommend making it part of the typical install? As even on a CPU, it still works, just slower. But right now without the SAM flag included, it doesn't get installed, but is still imported in SAM.py and Interactive_SAM.py., throwing an error. Submitting a PR right now on devel in case you want it.

Maybe in the far future there could be a dialog box somewhere to set the model so if someone wants to use vit_b, instead of vit_h?

[maxcorsini]

Now SAM is an option because the vit_h is really demanding. What is wrong here is that if segment_everything is not installed the tools need to be removed from the toolbar to avoid issues. This is checked using importlib but perhaps there is some bug elsewhere, I will check ASAP.

[maxcorsini]

For the moment, install.py force SAM installation, when the bug will be found we came back to the optional installation of SAM.