Capturing Reality releases RealityCapture
Originally posted on 28 December 2015. Scroll down for news of the commercial release.
Slovakian startup Capturing Reality has made RealityCapture available in open beta. The promising new application generates 3D models of real-world objects from laser scan data and source photographs.
The software is capable of tackling very large mixed data sets, generating accurate triangle-based meshes of objects ranging from buildings to people, even on laptops and low-end hardware.
Converts large mixed data sets into clean production assets
RealityCapture is capable of accepting source data in a range of formats, including laser scans, DSLR photographs – both those from capture arrays and unstructured image sets – and even drone photography.
The software can align the data sets automatically, or users can oversee the process manually, and to judge from early user feedback, the quality of the meshes it creates is very good.
The video above shows raw data recorded by 3D scanning specialist Ten 24 using its multi-DSLR capture stage and RealityCapture. The meshes were decimated to 1 million polygons, but haven’t been cleaned up manually.
You can see textured renders from the project, which was an in-house tech test, on Ten 24’s website. (If you follow our Facebook feed, you may have seen some of them already.)
According to Ten 24, RealityCapture “pushes the limits of what can be achieved with photogrammetry. Super-fast and ultra detailed, [it] is the critical software breakthrough that has been needed to take things to the next level.”
Process tens of thousands of source images
As well as generating accurate meshes and textures, RealityCapture is capable of handling very large data sets.
The software accepts an unlimited number of source images, and even an unlimited number of source point clouds, and is capable of using multiple available CPU and GPU cores on a local network to process the data.
According to the developer, a Core i7 machine with 32GB RAM can process 1,000 laser scans and 10,000 source images in “less than a week”.
And while it’s unlikely that many people will try, the firm’s other use case – processing 500 12 megapixel images on a notebook in the field – is far more standard, and is estimated to take “less than 10 minutes”.
Try it for free until 31 January 2016
The product website lists other interesting features, including georeferencing and automatic filtering of moving objects from source images, but there isn’t a lot of documentation, beyond the introductory video tutorials.
Models can be exported in PLY or OBJ format, complete with textures.
Updated 4 February 2016: RealityCapture has just been released commercially. The price is pretty high: a perpetual licence costs €15,000 (around $16,735), or €7,500 ($8,370) for the command-line version.
If that’s out of your range, there is a rather more affordable rental-only option for freelancers, which costs €99 ($99) for three months, including free updates, but without tech support and capping projects at 2,500 images.
The software is available for 64-bit Windows 7 and above. You’ll need a Nvidia GPU with CUDA 2.0 or higher.
Read more about RealityCapture on Capturing Reality’s website
Updated 5 February: The original version of this story listed a currency conversion for RealityCapture’s three-month rental option. We’ve now updated this to the actual dollar price.