Most image-based modelling packages can recreate 3D models from two or more reference photos of an object. 3-Sweep goes one better – or rather, one fewer.
The research project, developed by Tao Chen, Zhe Zhu, Ariel Shamir, Shi-Min Hu and Daniel Cohen-Or, and due to be presented at Siggraph Asia next week, recreates 3D models from a single image.
Three drags and you’re done
To help 3-Sweep generate the model, the user drags on the image to define the object’s profile, then once more to drag the profile along the object’s main axis.
The resulting textured 3D object can then be manipulated directly within the image, with the user rotating, scaling and editing the model as they would within 3D software. 3-Sweep automatically fills in the background.
The demo video (above) was posted on YouTube last month, but suddenly went viral this week – and with good reason: it’s one of the most eye-catching tech demos we’ve seen in a long while.
The limitations of the system are pretty clear – it only works on hard-surface objects with a sweepable profile, for a start, so you won’t be repositioning any people; and it struggles with concave or transparent surfaces – but the speed with which objects can be created and repositioned is extraordinary.