Define the registry for ML Backend Services & Models

[mihow]

The user should be able to see all available all available processing pipelines to choose from in the AMI Platform, and choose an appropriate one for their project.

• What ML Service Backends are available & online?
• What pre-configured Pipelines does each service have available? What applications are they good for?
• What models/algorithms does each Pipeline use? What version, how many classes, and other metadata.
• Georegion (polygon or rough bbox) that each model is applicable to. Populate pipeline choices based on deployment/project locations.
• List of species in training set
• Other metrics?

Some pipelines are available to all projects, some are available to a single project only (private ML backend service)

TODO

• Diagram our existing architecture
• Diagram where we want to go
• Explore Weaver for orchestrating the selection ML services
• Explore PySpark, Prefect, other relevant options

Relevant existing UI components

[mihow]

Notes from today's discussion: Can we link directly to this model registry from the landing page? To show what models the platform has and details about number of species each can recognize.

[adityajain07]

Having a model registry or simply a list of models available for public view. Add details like:

• Subheadings for localization, classification and tracking models
• Whether in-house or contributed by another group; have a reference link to paper for the latter
• The number of classes available and regions served