This paper introduces a novel dataset designed to advance automated analysis of dog facial expressions, which is crucial for understanding animal internal states like pain and emotions. Developed in response to a significant shortage of animal facial landmark datasets compared to human-related studies, and inspired by the CatFLW dataset, DogFLW comprises 3,274 annotated images of dogs, each marked with 46 facial anatomy-based landmarks. The landmark scheme was meticulously defined by three DogFACS-certified researchers, grounded in dog facial anatomy and the DogFACS method, and the images were sourced from the Stanford Dog dataset, filtered to include single dog faces in non-laboratory conditions. Each image is also annotated with a face bounding box and includes a visibility indicator for occluded landmarks. Benchmarks using the Ensemble Landmark Detector (ELD) and YOLOv8 demonstrated that while detection accuracy improves with dataset size, challenges remain particularly with ear landmarks due to the diverse shapes and lengths across breeds, and with breeds exhibiting long fur or obscured facial features. The authors hope this dataset will significantly aid the scientific community in utilizing AI-driven methods to deepen the understanding of dog behavior and emotional welfare, with future work focusing on enriching the dataset and classifying internal states. The DogFLW dataset is available upon reasonable request from the corresponding author.
Non-Invasive Computer Vision-Based Fruit Fly Larvae Differentiation: Ceratitis capitata and Bactrocera zonata
This paper proposes a novel, non-invasive method using computer vision
