Face Alignment Using Regression Trees

An implementation of Kazemi et al.’s One Millisecond Face Alignment With An Ensemble of Regression Trees.

Results so far:

The detector starts with a mean face shape (shown in black) and outputs a predicted face shape (shown in white). These images were not in the training set.

Result 1 Result 2 Result 3 Result 4

The pairs chosen for the node splits over all regression trees:

visualization

Pairs from an earlier detector mapped onto a face:

early visualization