Lightweight anomaly detection via knowledge distillation from ensemble teachers

This repository contains the code for the paper "Lightweight Anomaly Detection via Knowledge Distillation from Ensemble Teachers". The method uses a heterogeneous ensemble of Isolation Forests as the teacher and a lightweight MLP as the student. Knowledge is transferred via Kullback–Leibler (KL) divergence minimization. The code supports feature extraction, teacher ensemble construction, student distillation, and evaluation on the MVTec 3D-AD dataset.