The advances in Computer Vision are becoming more and more impressive. The suite I know best and with use more is Azure Cognitive services, however, there are surprises and advances that leave me with my mouth open.
This is the case of the work published by Aaron S. Jackson, Adrian encyclical, Vasileios Argyriou and Georgios Tzimiropoulos, where he explains how he can create a 3d model from a 2d photo. The best thing is to see it in action
I recommend you also see the video where they apply the algorithm in real-time to faces in a video.
Now is the time to try to explain, with my words of a 5-year-old boy, how this works. Behind this algorithm is a Convolutional Neural Network (CNN), which has been trained with 2D images with the results expected in 3D. The interesting thing about this model is that it has reached such a level of sophistication that it does not need a specific point of reference for a face, it works on any face.
With the 2D image information, it is possible to rebuild elements of the face, including parts that are not seen in the 2D image. In this way, and after much training CNN, achieve the results that can be seen in the live demo!
Maybe it’s better to hear this in their own words
Greetings @ Toronto
- Aaron S. Jackson, Adrian Bulat, Vasileios Argyriou and Georgios Tzimiropoulos, Large Pose 3D Face Reconstruction from a Single Image via Direct Volumetric CNN Regression
- Live demo
- Github code
- Cognitive Services, Computer Vision
- Cognitive Services, Custom Vision Services