Coarse facial feature detection in sheep
Lech Szymanski, Michael Lee · 2021
We present a deep learning model for facial feature detection in sheep, which is part of a project on machine learning for kinship detection in livestock. Using YOLO-like training, we obtained a model capable of identifying the eye and nose line of the animal in the image, with a mean test error of ∼5 pixels (in a 256×384 input image). We further evaluated the effectiveness of the resulting pose estimation and alignment for kinship prediction and report its balanced accuracy of 73% – an improvement of 5% over our previous effort that had no benefit of the image pre-processing provided by the proposed feature detection network.