Transfer Learning in Animal Facial Expression Recognition: A Comparative Study Using Human Pre-Trained Models
Year: 2024
Type: conference-paper
Abstract: This study explores techniques that enhance the efficacy and dependability of animal facial expression recognition systems, especially for cats and dogs, under the constraint of small-scale annotated datasets. The paper examines two approaches to the animal facial expression recognition task, specifically focusing on cats and dogs. The first approach involves training a model on a limited dataset of animal facial images. The second approach utilizes transfer learning with a model developed for human expression recognition, which is then fine-tuned for animal facial expression recognition. This research aims to establish that given scarce labeled animal datasets, transfer learning refines the generalization capabilities of animal expression recognition systems. The results indicate that transfer learning is beneficial. However, the small size of the animal dataset and the original accuracy of the human-centric model diminish its impact. This study demonstrates the feasibility of applying transfer learning, while underscoring the need for larger datasets and the further refinement of human facial emotion models to strengthen animal facial expression models with high accuracy and dependability.
Author: Emily G. Wang
Institution: University of California, Irvine
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4412524843