Fig. 3.

Fitting examples of pixel-based AAMs from the test set of LFPW. First row: POIC. Second row: aFast-SIC. Third row: E-POIC. POIC does not perform well, however aFast-SIC and E-POIC achieved satisfactory robustness and accuracy in landmark localization. Notably, to obtain these results, the appearance model of the AAMs was built using raw un-normalized pixel intensities as features. Neither sophisticated shape priors or robust norms were used during fitting nor robust image features were employed to build the AAMs: we simply trained the AAMs in-the-wild on the same database