Can Facial Landmarks Guide us to Understanding Animal Emotions?

George Martvel#, Anna Zamansky & Ilan Shimshoni
Department of Information Systems

 Artificial Intelligence
Computer Vision

Veterinary Medicine

PhD Grant 2023

According to A. Krause and C. Nawroth from Leibniz-Institute for Farm Animal Biology, the facial expressions of animals usually look different than those of humans — joy might not be indicated by a smile (showing your teeth is often a signal of stress in other primates), while sadness is not accompanied by tears (pigs do not cry). But how then do we correlate between emotions and facial expressions in various animals?

In our research in the Tech4Animals lab, we aim to develop an automatic system based on computer vision networks that allow the detection of facial landmarks for various animals. Using these facial landmarks, it becomes possible to determine emotions and pain in animals by non-invasive methods.

In my PhD, I study the complex computational problem of automated detection of facial landmarks for various animals using state of the art computer vision techniques. I am developing a novel approach based on an ensemble magnifying method, allowing focusing on parts of animal faces, which makes landmark detection more precise and accurate. I have already developed an accurate model for 48 landmarks of cat faces; its generalization to humans also performs well comparable to the state-of-the-art models for humans; a model for dogs is currently in development. These results enable better and more explainable automated recognition of emotions and pain in animals by non-invasive methods.
Future work includes expanding this system for various animals (sheep, monkeys, etc.) and finding more applications for the automated facial landmark detection models.

Links to paper and dataset:

https://arxiv.org/abs/2305.04232

https://www.tech4animals.org/catflw