Add support for onnx/yolo detection models to complement build in AI Segmentation

[Diemsgit]

This enhanced creation is just a magnificent piece of engineering ! Thank you so much for this more than wonderful solution.

As the current version already supports so many plugins and functionalities out of the box which includes AI segmentation, it would be more than welcome to complement this with custom detection (yolo/onnx) models supporting multiple classes. Any detection classes can be created as layers.

[giswqs]

@Diemsgit Thank you for the kind words. sam3-onnx will be added soon.

References:

• https://huggingface.co/vietanhdev/segment-anything-3-onnx-models
• https://github.com/wkentaro/sam3-onnx
• https://github.com/vietanhdev/samexporter
• https://www.npmjs.com/package/onnxruntime-web

[Diemsgit]

Thank you!
Looking very forward to any future updates.

[Diemsgit]

I feel it's almost disrespectful to ask if there is any progress for this feature as your working too hard with all the new releases ..
Think it would suffice to just be able to load an offline (local) onnx model file to appy on layer of choice. This then creates extra layers for the classes detected.

[giswqs]

Implemented in #905. It is for demonstration purpose only. None of the ONNX models work well for remote sensing imagery. That's the reason I have been hesitant to add it to GeoLibre. For real GeoAI work, I would suggest using the GeoAI package.