Abstract
Digitization of anatomical landmarks is a critical process in geometric morphometrics, challenged by the labor-intensive, and error-prone nature of manual methods. Application of machine learning or AI-automation is accordingly highly anticipated, but only a few publications have so far reported success in this area and offered suitable software.Such an application would be highly beneficial, particularly when dealing with abundant material. It is recommended that users (human) repeat their manual digitizing at least once to verify their own precision. User error is generally less than the inter-user error, i.e., the discrepancies observed when different operators digitize the same objects. This inter-user error is generally high, preventing data digitized by different operators from being used together.The use of automated digitization by AI is still in its infancy and requires a clear answer to a few questions. Two of them are:- Do different AI-based software produce differences when digitizing the same objects?- Is AI-based software built for species A effective for digitizing species B?We present a first analysis which provides elements of answer to these two questions. Our methodology involves the repeated digitization of a diverse set of landmarks on multiple species by both AI algorithms and human experts, measuring precision and inter-user variability.