Hi Sejin,
Thank you for your questions and your detailed post.
A1: Great question. You can download Sreet View images in many unique ways by creating a custom pipeline using FootprintHandler and ImHandler. Do you have a specific method in mind, such as using a list of addresses, etc.?
A2: I think segmentation results should always be post-processed before attribute extraction. For example, if the algorithm detects multiple doors, you can eliminate the smaller detection assuming there will always be one front door to a single-family building. As for windows, polygonization using edge detection could be another method. These ideas will eventually get implemented into BRAILS, but for now, what is in FacadeParser is what BRAILS currently offers.
A3: We have a prototype idea currently implemented in BRAILS based on limiting the field of view to just the building of interest. However, this feature is still undergoing testing and may not be released for another month or two.
I hope this helps!