General Questions

[drw02]

Dear,

Can you kindly direct me to where i can find how each classification is defined. For example, how is a structure classified as having soft story?
Moreover, is there any available documentation on how the street-view images were transformed into normal 2d images containing only one building. Was this process autonomous, or was it manually done?

I highly appreciate your help!
Thank you in advance for you time.

[fmk]

Can you be a bit more specific, for example are you looking at the different individual BRAILS modules or are you using CityBuilder.  You might want to look at day 1 of the material presented for a recent workshop that SimCenter gave in which image classification is discussed:

https://nheri-simcenter.github.io/SimCenterAI_Workshop2021/source/lecture_videos.html#day-1-introduction-to-machine-learning-part1-2-and-part3-and-brails

There is no manual processing of images to create images used in CityBuilder when they are downloaded from Google. What a user using the individual modules does to their images before they use any of the modules in BRAILS is up to the user; we don't do any manual processing.

[drw02]

Thank you for your reply!

To be more specific when using the soft story BRAILS modules for example, how is the soft story defined? Does the module detect the height difference between the first story and the rest of the building?
My main concern is how does the model define a soft story, is it provided in the documentation or somewhere else that you can refer me to?

Thank you again for your assistance!

[fmk]

Thanks for the question. You point out a weakness in the documentation which I will work to get improved.

The soft story model is an image classifier that utilizes a convolution neural network (CNN). Like the roof classifier and occupancy classifier, the training and test sets for the CNN used a collection of labelled images. The image sets for the soft story classifier consists of a set of street-view photos of buildings that were labelled as either being a soft story or not a soft story (with many of the labels being obtained from lists of weak story buildings obtained online, e.g. Santa Monica (https://www.arcgis.com/apps/webappviewer/index.html?id=05191306d93d4c04827773b8d2151cd7), Berkeley (https://www.cityofberkeley.info/softstory/) and others). 

For a more generic under standing of CNN's use in image classification see: https://developers.google.com/machine-learning/practica/image-classification

Better approaches to detecting soft stories are indeed possible that could include for example detecting floor levels and performing calculations as you suggest.

[drw02]

Thank you again for your reply!

I have a two more questions if you don't mind.

1. The input to the BRAILS modules (other then the roof classifier) are they a street view image meaning a panoramic picture or a regular rectilinear image only containing the building of interest?
2. How is the elevated foundation defined? By elevated you mean that part of the foundation is above soil level?

Thank you in advance!

[fmk]

1. yes a street view like image
2. it is a binary classifier that was trained using images of buildings on stilts versus buildings on the ground, see sample images here: https://nheri-simcenter.github.io/BRAILS-Documentation/common/user_manual/modules/foundationElevationClassifier.html

now we do have a module about to be released (October 2021) that provides a calculation of height of first floor level above the soil that goes beyond the image classifiers. It is part of BRAILS being developed at UCLA to determine needed info (roof pitch, roof elevation, first floor level). We have used the software in our testbeds to provide this information for our regional inventories.