Open mvsc1 opened 1 year ago
@mvsc1
This is indeed strange that PyTorch would be working prior to the configure tool and im not sure the source of this error. The configure tool here simply tries to utilize pip install samgeo which will install the SAM module directly from here.
Im not aware of any compatability issues between PyTorch and the underlying QGIS packages. Did you still have this issue?
Windows 10 QGIS 3.30.3 Python 3.9.13
Hello,
I just installed it using through the QGIS app and I get this as well. Anything you want me to look at or change ?
Thank you so much for the great work.
Windows 10 QGIS 3.30.3 Python 3.10.11
I'm also having this issue. Would love to see this working as intended. Looks dope.
Hi all,
I am looking into the issue. Seems to be some conflict with the libraries but im not sure which yet.
Unfortunately, I have not found a solution here. Any contributions/help related to this issue would be appreciated. Please keep in mind that the GPU will only be available for CUDA compatible devices.
One solution would be to create a bat file linked to anaconda for execution of the samgeo module to keep environments seperate.
Does a force reinstall of torch solve the issue here? i.e. from the OSGeo4W console type
pip3 install --force-reinstall torch
Windows 11 Python 3.9.12
Segment Anything Model (SAM) runs without GPU/cuda after being installed with Configure Segment Anything Model (SAM) in QGIS. It seems the configure tool installs its own torch 2.01 version which only runs on the CPU.
Before installation: PyTorch 2.0.1 with CUDA 11.8 installed - torch.cuda.is_available() returns True After installation: Another PyTorch 2.0.1 (without CUDA) is also installed - torch.cuda.is_available() returns False
If I uninstall torch 2.0.1 but keep torch 2.0.1 with cu118, the tool doesn't run anymore.
Any way to use the model with GPU in QGIS?