Vast sets of geometric training data for machine learning and AI – with an unprecedented amount of architectural resolution and metadata
Get in touch
To date, data sets of architectural spaces could be either extensive but of low credibility and resolution, or qualitative but requiring a too immense amount of time to reasonably compile manually. As a result, adequate training sets simply did not exist. Until now.
As architects ourselves we are heavily invested in the principles that make up various spaces. For example, an apartment would typically be arranged in such a way that the living space and bedrooms would gain façade access in favor of the hallway, bathrooms and storage.
Based purely on architectural logic, each ruleset respectively has been encoded to parametrically generate the geometries. As a result, Spaces by Parametric are extensively detailed, qualitative and fully architecturally believable.
Spaces by Parametric consists of geometry that has all been tagged with corresponding labels of what it is and where it is.
This base configuration is possible to complement with additional metadata, e.g. custom point clouds, should there be need to tailor simulations and trainings further. Be it sensor technologies or light optimization – Parametric Geometry Database provides reliable, customizable training data.
As more typologies get added and refinement continues, they do so in a pilot collaboration with Treble Technologies – an Icelandic acoustics firm pioneering wave-based simulations in the field.
Below are the labelled datasets currently available, with many more coming soon.
From phone booth to board room.
From student dorm to New York penthouse.
From ramen-place to fancy fine dining.
If you neither have units, blocks nor volumes you can still explore generative design. In Hektar Pro you can explore the Hektar default library.
