Automated Multi-View Silhouette Generation and Morphometric Clustering of Isolated Theropod Teeth Using Machine Learning - Code, data and models

Supplementary data and code for “Automated Multi-View Silhouette Generation and Morphometric Clustering of Isolated Theropod Teeth Using Machine Learning” This repository contains the supplementary code, data files, and other supporting material associated with the manuscript: Marques, C. S., Dufourq, E., Pereira, S., Santos, V. F., & Malafaia, E. Automated Multi-View Silhouette Generation and Morphometric Clustering of Isolated Theropod Teeth Using Machine Learning. The study presents an automated computer-vision and machine-learning workflow for generating standardized multi-view binary silhouettes of isolated theropod teeth, extracting morphometric measurements, exploring morphometric structure with UMAP and HDBSCAN, interpreting cluster-derived groups with random forest models, and evaluating visual recoverability of the groups using a multi-branch EfficientNet model with Grad-CAM visualizations. The repository includes cleaned Python scripts used for: generating synthetic training data; training and applying the DeepLabv3+ segmentation model; post-processing binary masks; extracting view- and region-based morphometric measurements; performing UMAP–HDBSCAN clustering and sensitivity analysis; interpreting cluster-derived groups with random forest classification and feature importance; training a multi-view EfficientNet classifier using repeated stratified cross-validation; applying the trained Random Forest and EfficientNet models to unseen teeth; comparing classifier predictions for unseen specimens. The repository also includes the different datasets, output tables and images, and trained models.. Directory are generic folder paths and users should edit the path variables at the beginning of each script to match their own local directory structure before running the workflow.   The repository contains the following folders: Data folder: Contains the images and tables needed for training the models and doing the analysis.  Code folder: Contains the code used to do all the analysis. Each file contains a different step of the analysis. Models folder: Contains the trained models obtained by each step of the analysis. Results folder: Contains additional images obtained during the analysis.