3D scanning technologies have been constantly improving and will continue to improve as time goes on; but one of the latest advancements in 3D has come out of Kingston University and the University of Nottingham. Researchers there have opened the scope of 3D scanning from faces being scanned with a 3D model result to 2D flat image to 3D reconstruction model.
The way they achieved this is by using a Convolutional Neural Network (CNN) to use image pixels to map the 3D data. Researchers from both universities published a paper with their findings and discuss more about their processes: “Large Pose 3D Face Reconstruction from a Single Image via Direct Volumetric CNN Regression.”
Facial Scanning with a CNN
Usually to extract 3D data from 2D images, there needs to be many different shots at different angles. “Besides its simplicity, our approach works with totally unconstrained images downloaded from the web, including facial images of arbitrary poses, facial expressions and occlusions.”
Using the CNN multiple facial expressions from multiple angles the system can now overcome the challenge of illumination and irregular facial positions. The CNN is able to fill in the gaps and guess (with an incredible amount of accuracy) the non visible parts of the face on a 2D image or video and reconstruct the entire face in 3D.
The team has made an online demo tool to 3D reconstruct 2D faces of your own, here. Code and models will soon be availiable as well. With the online demo tool, just upload your image and let the CNN do its magic. In less than a minute, the virtual model is created. Within 20 minutes the image and model will be deleted. According to the demo page, 300,378 faces have been uploaded since early September 2017.
This CNN software has a lot of creative applications for games, social interaction, VR and even security and law enforcement. To lear more about the project, click here. To lear more about our 3D scanning services, click here. For any and all of your 3D scanning needs, Arrival 3D has you covered.
