Brails

What's Changed
- TranspInventoryGenerator now generated graph networks from TIGER data
- Miscellaneous performance updates to existing building prediction capabilities

Miscellaneous stability improvements and fixes to inventory generation and data scraping modules

Major Changes:
- Enabled rapidtools feature to extract building-level imagery from NHERI RAPID aerial and street-level image data
- Added the capability to pull National Structure Inventory as baseline building inventory
- Updated TranspInventoryGenerator for more stable API queries
- Updated FacadeParser for more effective memory utilization

Major Changes:

- Enabled Transportation inventory generation for roadways, bridges, tunnels, and railroads
- Revised FacadeParser that calculates dimensions from depth map as opposed to camera coordinates for more robust dimension predictions
- Added FEMA USA Structures as an additional building footprint source

BRAILS version 3.0.0 contains a new interface that simplifies inventory generation and training image classification/semantic segmentation models.

In terms of inventory generation capabilities, BRAILS now contains modules that can predict: 1) roof type, 2) roof cover type, 3) occupancy class, 4) era of construction, 5) the number of floors, 6) building height, 7) roof eave height, 8) roof pitch, 9) facade window area, 10) first-floor height, 11) existence of chimneys, and 12) existence of garages.

The new InventoryGenerator workflow offers a convenient end-to-end workflow to create building inventories. InventoryGenerator
1) Extracts location polynomials using Nominatim API,
2) Parses footprint information from Microsoft Footprint Database or OpenStreetMaps,
3) Downloads Google street-level or satellite imagery explicitly focused on each building in the region specified by the user,
4) Passes these images through BRAILS modules, and
5) Prints out simple CSV files for ingestion into many other platforms, such as R2D or QGIS.

All of this can now be performed in a few lines of code.

BRAILS now also offers automated pipelines for training image classification and semantic segmentation models. Using ImageClassifier and ImageSegmenter, it is possible to train state-of-the-art deep learning models without needing to spend developing time on the training step of the process.

BRAILS version 2.0 is re-structured with modules for performing specific analyses of images. The expanded module library enables BRAILS’ capability of predicting a larger spectrum of building attributes including occupancy type, roof type, foundation elevation, year built, soft-story, number of floors.

The new version also features a streamlined workflow, CityBuilder, for automatic creation of regional-scale building inventories by fusing multiple sources of data such as OpenStreetMap, Microsoft Footprint Data, and Google Maps, and extracting information from them using the modules.