Evaluation protocol used in the referenced dairy-cow behavior study

Determine the evaluation protocol used in the referenced study [21] on vision-based dairy-cow behavior detection, which is not specified in the discussion and is needed to assess whether its reported contraction-detection accuracy is comparable with the results of the Slow R50 and Multiscale Vision Transformer models.

Background

The paper compares its contraction-detection results with those reported in study [21]. The authors note that the referenced study used a substantially smaller number of contraction actions and apparently involved relatively few cows exhibiting contractions, making the robustness and comparability of its reported accuracy uncertain.

The authors explicitly state that they do not know the evaluation protocol used in study [21]. Establishing that protocol would clarify whether the reported results were obtained under a methodology comparable to the paper’s box-based train/test split and whether the accuracy values can be meaningfully compared.