Wolfram Language

Improved Face Detection

Face detection is the base operation for several other face analysis and transformation operations, including face recognition, pose estimation, facial keypoint, morphing and more. FindFaces has been significantly improved to detect faces faster and more robustly, with fewer false positives.

Comparing quality assessment for Versions 11 and 12.

The new face detection is also stronger at detecting rotated faces.

Here is a benchmark of FindFaces computed on the FDDB and CelebA datasets, which shows a 30% increase in quality.

* AP@0.4 means that the intersection-of-union (IoU) detection threshold of 0.4 is used.

In addition, the average timing of the FindFaces evaluation is also improved. The following benchmark shows the speed improvement for images of different sizes.

* Tested on an Intel core TM i7 7500@2.7Ghz laptop.

Related Examples