Evaluation of Deep Learning Instance Segmentation Models for Pig Precision Livestock Farming

In this paper, the deep learning instance segmentation architectures DetectoRS, SOLOv2, DETR and Mask R-CNN were applied to data from field of Pig Precision Livestock Farming investigate whether these models can address specific challenges domain. For purpose, we created a custom dataset consisting 731 images with high heterogeneity high-quality masks. evaluation, standard metric for benchmarking in computer vision, mean average precision, was used. The results show that all tested be considered domain terms prediction accuracy. With mAP 0.848, DetectoRS achieves best on test set, but is also largest model greatest hardware requirements. It turns out increasing complexity size does not have large impact accuracy pigs. DETR, achieve similar parameter count almost three times smaller. Visual evaluation predictions shows quality differences generates masks overall, while has advantages correctly segmenting tail region. However, it observed each problems assigning once pig overlapped. demonstrate potential lay foundation future research area.