Hacker News Reader

I think the breakthrough with Facebook's system was the idea to transform the face before running the training or verification on it. And the new system seems to use the same idea: The new algorithm works by normalising each face into a 150 x 120 pixel image, by transforming it based on five image landmarks: the position of both eyes, the nose and the two corners of the mouth.

View on HN

No. Alignment is done by every face verification method, and even the "new" 3d alignment done here is on par with other alignment methods in practice today.

At their talk at CVPR, they made clear that the biggest factor in getting their numbers was having tons of training data and a high-capacity learning model. The paper even shows that if you replace their 3d alignment with a standard 2d affine alignment (something people were doing 15 years ago), you lose only 1.5% on LFW (table 2).

Reply on news.ycombinator.com