Building Inventory Attribute Extraction

• 

  3D projection of Building Using 2D image

• 

  Segmented GAN output of a building

Inspiration

We need to develop AI solution for climate change

What it does

Our System helps fill the gap of

How we built it

1. We produced annotation data for 3D models of building.
2. Trained an ensemble model for Building Landmarks.
3. Searched for segmentation dataset of buildings.
4. Trained dualGAN on the dataset and infer it on MONTREAL buildings.

Challenges we ran into

1. Post processing the segmentation mask to extract the windows data.
2. Post processing the 3D output of our models to Height using a regression model with building footprint data as well.

Accomplishments that we're proud of

We were able to train 2 models.

1. Provides segmented mask of building which will be used further to produce number of windows, their locationd and wall to window ratio.
2. The other ensemble tree model gives 3D projection of buildings on 2D images.

What we learned

Application of machine learning solutions to real application

What's next for Building Attributes

Create a scalable solution which updates and improvise with more data

Built With

• dlib
• gans
• python