Document : Design Choices

[kshitijrajsharma]

• Document the design choices including :
• UI text choices
• Git approach for the model contribution
• infrastructure choices !
• Licenses ( For images , cc by 4.0 , for code : AGPL v3 , for data ( most likely ODBL ) , to be decided .... ) , We might support custom license as well there should be list of the licenses supported & should be documented
• Input data ( High resolution aerial imagery most likely 256/256 or 512 / 512 , along with labels as geojson ( point line and polygon )
• Category ( We will support semantic segmentation , instance segmentation & object detection )
• Model output ( Point , line or polygon only ) , if some models only add attributes like road classification then it should take geojson as input classify add attribute and spit the geojson back to keep the input output constant
• and more

We keep the discussion here in issue and then we can raise PR to include all of this in docs !

[kshitijrajsharma]

We want to keep as simple as it is for the user , this would be the overall userflow

We opted for this git approach primarily because :

• It allows developers to easily contribute the model
• CI/CD will help us to test the model , if it is working or not , check docker builds , check compatibility with our STAC
• Allows seamless review and updates
• No need to setup any custom form on UI , completely removes the UI Frotnend Design along with complex database
• It is most friendly for version control to stage / prod directly from the github

[kshitijrajsharma]

Todos :
@AbdelrahmanKatkat to document why we opted for the UI texts , what will be communicated to the developers / anyone who wants to contribute
@kshitijrajsharma to document the zenml choices , add reference (https://github.com/hotosm/fAIr/blob/develop/docs/decisions/infra/0001-mlops.md )