15,000 Skeletons, or How I Learned to Stop Worrying and Love the Phenotype
Description
Conference talk (Macroevolution V) presenting a computer vision workflow for large-scale digitization of avian skeletal morphology. Describes use of convolutional neural networks (including U-Net and Mask R-CNN approaches) to extract linear measurements from museum specimens with low error (reported RMSE ~0.89 mm). Introduces the Skelevision pipeline and outlines integration with multivariate phylogenetic models to detect shifts in high-dimensional phenotypic evolution across passerine radiations.