Since there is no 3D building dataset for Deep Learning in architecture, we decided to make a framework that would generate synthetic 3D buildings suitable for DL tasks.

Our framework generates different data types: mesh, point cloud, rendered images, segmentation masks, depth and surface normals annotations. We made it modular and extendable, so the users can add as many details as they need of the type they need. 

This research was featured by Deep AI among trending works in Deep Learning in May 2021.

​

Frameworks: Blender, python

​

Year: 2021

Location: @Milan @CDInstitute

Collaborators: Alberto Tono, Meher Shashwat Nigam, Cecilia Bolognesi

​

Video    Github     Arxiv    DeepAI     Project website

​

Inspiration: IKEA dataset.

​

How to cite: 

Fedorova, S., Tono, A., Nigam, M. S., Zhang, J., Ahmadnia, A., Bolognesi, C., and Michels, D.: SYNTHETIC DATA GENERATION PIPELINE FOR GEOMETRIC DEEP LEARNING IN ARCHITECTURE, Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLIII-B2-2021, 337–344, https://doi.org/10.5194/isprs-archives-XLIII-B2-2021-337-2021, 2021.