AU Recognition

Facial action unit (AU) recognition is formulated as a supervised learning problem by recent works. However, the complex labeling …

Instead of minimizing the distance of two joint distributions directly, which requires the estimation of the marginal distribution of the input, the proposed approach uses an adversarial strategy to exploit the probabilistic duality, thus avoiding the estimation of marginal distribution.

We propose employing the latent regression Bayesian network to effectively capture the high-order and global dependencies among AUs.

we propose a novel generative adversarial recognition network (GARN) for personalized AU recognition without any assumptions. Specifically, the proposed GARN consists of a generator, a discriminator, and a classifier.

We propose a novel weakly supervised AU recognition method to learn AU classifiers with only expression labels. Specifically, we notice that there exist domain knowledge about expressions and AUs that can be represented as prior probabilities. We generate pseudo AU data for each expression; for AU classifiers’ training, we propose an RAN model, which consists of a recognition model and a discrimination mode trained simultaneously by leveraging an adversarial process, to make the distribution of the recognized AU close to the distribution of the pseudo AU data. Furthermore, we extend the proposed method to semi-supervised learning with partially AU-annotated images.

Inspired by the observations that AUs are samples of the underlying AU label distributions, we propose a deep facial action unit recognition approach learning from partially AU-labeled training data through incorporating such spatial regular patterns of AU labels presented in ground-truth AU labels into the learning process of AU classifiers from a large-scale facial images without AU annotations.