Import with optional dependencies

[guillemc23]

Hello @giswqs! Really solid work with this library!

I would like to propose being able to install this package in different ways, depending on the project's needs. For example, in my use case, I don't need a lot of the dependencies since I only need to use the model itself, but having to install leafmap adds a lot of overhead for Docker environments.

I propose using the pyproject.toml to define different optional dependencies:

• segment-geospatial[core]: only installs the actual model utilities
• segment-geospatial or segment-geospatial[full]: installs everything including leafmap, localtileserver and other utilities that are only required when running for Jupyter notebooks
• segment-geospatial[text]: installs SAMGeo with GroundingDino to enable text prompts

For example, one could do pip install segment-geospatial[core,text] if his or her use case does not require Jupyter notebooks and is instead planned to run as a service with some other frontend.

I think it would be interesting as well to provide options for whether to install torch with CPU or GPU support, since, at the moment, all of Nvidia's libraries are being installed even on CPU only machines, making the environment and Docker image unnecessarily big. I can suggest these, with one of them being the default installation mode.

• segment-geospatial[gpu]: installs torch from the CUDA repository
• segment-geospatial[cpu]: installs CPU only version

PD: I believe GroundingDino and SAMGeo should have their OpenCV version aligned, since the first requires opencv-python and the second one opencv-python-headless, leading to some import errors when installing both. This can be fixed overriding dependencies with the UV Python package manager, but that's just a local workaround

[giswqs]

Good suggestions! Could you please submit a pull request?