RealityPrint

• 

  Image 1 of the Warehouse

• 

  Image 2 of the Warehouse

• 

  Image 3 of the Warehouse

• 

  A 3D model automatically constructed as a point cloud from a set of photos of warehouses.

• 

  Warehouse 3D model from another perspective.

Guild: Noria - Quiet Hacking Space Table Number: G-3

Inspiration

Blender and other 3D modelling applications can be hard to master. You can spend hours trying to turn the default blender cube into an object - or you can use RealityPrint and turn that video into a 3D model (a point cloud) automatically. Inspired by the theme of Legacy, this technique creates a 3D model from a video with no AI - only using traditional algorithms.

What it does

RealityPrint allows you to turn any video into a 3D model using a technique known as Photogrammetry.

How we built it

A python library known as pycolmap.

Challenges we ran into

My laptop has an RTX 4060 laptop GPU, which is capable of using Nvidia Cuda. However despite this, it still takes roughly an hour for every 10 seconds of video you render, which is far from ideal.

Accomplishments that we're proud of

Ultimately, the project is a proof of concept, and I was able to turn a video into a point cloud which roughly resembles it, although with tweaks to turn it into a mesh and remove noise a necessary later step in the pipeline.

What we learned

I mainly learnt about implementing Photogrammetry.

What's next for RealityPrint

Currently the output is a point cloud - it is possible to turn it into a mesh, but the results won't necessarily resemble the object you want. Therefore, a more effective way of implementing the point cloud as a mesh is needed.

Built With

• open3d
• pycolmap
• python