This repository contains the training and evaluation code for the mosquito species classification model powering the Mosaico Webapp. The core is a convolutional neural network (CNN) with ~50 million parameters, fine-tuned on a private dataset collected by the Istituto Superiore di Sanità (ISS).

• Private dataset of mosquito images from across Italy and some international samples.
• Covers 15 distinct mosquito species.
• Data collected and annotated by expert entomologists.

• Base CNN model fine-tuned using Modified Evidential Deep Learning (EDL).
• Designed for Open Set Classification to handle unknown or novel classes.
• Provides calibrated uncertainty estimates on predictions, important for expert review and feedback loops.

• Training scripts with configurable hyperparameters.
• Evaluation and test pipelines producing metrics and uncertainty calibration analysis.
• Integration-ready checkpoints for deployment in the main Mosaico web platform.