Is SPARK capable of calibrating for geometric correction and screen blending automatically or do we need to match virtual 3D and physical sculpture manually by exporting accurate models from 3D package software like what is done in the 360 VR tutorial video?

As you suggested you have to do the semi automatic calibration process:

Ideal workflow:

1. Create a precise model (i.e. laserscan) of the object. clean up / simplyfy the mesh and create UV-coordinates.

2. Identify easy to distinguish features of your object (at least 5 or 6 visible features from each projector)

3. create a low polygon mesh with just the vertices that represent those features. too many vertices make the targeting of the calibration process cumbersome or even downright impossible because you are never sure if you have chosen the correct vertice – which would lead to an erroneous result.

4. stick small reflectors on the features and use a headlamp to see them when targeting. if you choose light grey ones you can even leave them on the model in case you need to calibrate again.

5. calibrate the projectors as shown in the mentioned tutorial.

6. if your 3d model was accurate and you didnt mistake any vertices you should get a very precise solution.

7. Once you have accurate projector positions in relation to your model, the softedges are calculated  automatically and don’t need any manual intervention.

8. use the MeshWarp node to make final adjustments if it doesn’t fit in some places. since the object is not moving in space, it is totally ok to fix things this way.