Feature : Multimasks training for RAMP

[kshitijrajsharma]

What does this PR do ?

API

• Adds API endpoint for training two support three new parameters input_contact_spacing , input_boundary_width and multimasks
• Provides option in training to choose between binary masks and multimasks
• Records multimasks extra param in description of training if multimasks is chosen

Frontend

• Adds option for user to enable multimask feature and provide input for boundary width and contact spacing

• Integrates https://github.com/hotosm/fAIr-utilities/pull/29

Screenshots

How to test ?

• Download preprocessed.tar.xz from training with multimasks enabled , You should have multimasks folder added

• When you load them in to QGIS you should see multiple bands for the masks like following . You can compare with the binary masks if you like

• While doing training you will notice multimasks folder being used if enabled.

• if multimask feature is disabled then things should work as it is as used to before

• Compare the prediction with and without the boundary option ( specially in dense area )