Dataset Information

Unmasking the Devil in the Details: What Works for Deep Facial Action Coding?

ABSTRACT: The performance of automated facial expression coding has improving steadily as evidenced by results of the latest Facial Expression Recognition and Analysis (FERA 2017) Challenge. Advances in deep learning techniques have been key to this success. Yet the contribution of critical design choices remains largely unknown. Using the FERA 2017 database, we systematically evaluated design choices in pre-training, feature alignment, model size selection, and optimizer details. Our findings vary from the counter-intuitive (e.g., generic pre-training outperformed face-specific models) to best practices in tuning optimizers. Informed by what we found, we developed an architecture that exceeded state-of-the-art on FERA 2017. We achieved a 3.5% increase in F1 score for occurrence detection and a 5.8% increase in ICC for intensity estimation.

SUBMITTER: Niinuma K

PROVIDER: S-EPMC7274256 | biostudies-literature | 2019 Sep

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Unmasking the Devil in the Details: What Works for Deep Facial Action Coding?

Niinuma Koichiro K Jeni Laszlo A LA Ertugrul Itir Onal IO Cohn Jeffrey F JF

BMVC : proceedings of the British Machine Vision Conference. British Machine Vision Conference 20190901

The performance of automated facial expression coding has improving steadily as evidenced by results of the latest Facial Expression Recognition and Analysis (FERA 2017) Challenge. Advances in deep learning techniques have been key to this success. Yet the contribution of critical design choices remains largely unknown. Using the FERA 2017 database, we systematically evaluated design choices in pre-training, feature alignment, model size selection, and optimizer details. Our findings vary from t ...[more]

PMID: 32510058

Similar Datasets

Project description:Facial expressions are subtle cues, central for communication and conveying emotions in mammals. Traditionally, facial expressions have been classified as a whole (e.g. happy, angry, bared-teeth), due to automatic face processing in the human brain, i.e., humans categorise emotions globally, but are not aware of subtle or isolated cues such as an eyebrow raise. Moreover, the same facial configuration (e.g. lip corners pulled backwards exposing teeth) can convey widely different information depending on the species (e.g. humans: happiness; chimpanzees: fear). The Facial Action Coding System (FACS) is considered the gold standard for investigating human facial behaviour and avoids subjective interpretations of meaning by objectively measuring independent movements linked to facial muscles, called Action Units (AUs). Following a similar methodology, we developed the CalliFACS for the common marmoset. First, we determined the facial muscular plan of the common marmoset by examining dissections from the literature. Second, we recorded common marmosets in a variety of contexts (e.g. grooming, feeding, play, human interaction, veterinary procedures), and selected clips from online databases (e.g. YouTube) to identify their facial movements. Individual facial movements were classified according to appearance changes produced by the corresponding underlying musculature. A diverse repertoire of 33 facial movements was identified in the common marmoset (15 Action Units, 15 Action Descriptors and 3 Ear Action Descriptors). Although we observed a reduced range of facial movement when compared to the HumanFACS, the common marmoset's range of facial movements was larger than predicted according to their socio-ecology and facial morphology, which indicates their importance for social interactions. CalliFACS is a scientific tool to measure facial movements, and thus, allows us to better understand the common marmoset's expressions and communication. As common marmosets have become increasingly popular laboratory animal models, from neuroscience to cognition, CalliFACS can be used as an important tool to evaluate their welfare, particularly in captivity.

Dataset Information

Unmasking the Devil in the Details: What Works for Deep Facial Action Coding?

Publications

Unmasking the Devil in the Details: What Works for Deep Facial Action Coding?

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets