Use sparse view image as input for inference

[kenchen3000]

Thanks for sharing and great work! from the readme it seems that run.py can render selected scene. I am wondering how to generate and render novel scene views from a sparse set of the images as mentioned in paper? thanks.