DualSOM: Dual-mode software framework for clustering and classification using self-organising maps and sparse autoencoders

Le résumé fourni par la source

This work presents an open-source software framework for unsupervised clustering and supervised classification of high-dimensional data, demonstrated on human posture recognition from RGB-D skeletal inputs. The software combines sparse autoencoding for dimensionality reduction with a self-organising map trained using distance-based learning. A unified dual-mode pipeline supports both automatic clustering and label-based classification without changes to the model structure. The framework is modular, reproducible, and suitable for real-time operation, with particular relevance to robotic perception systems. Validation on public and proprietary datasets demonstrates competitive clustering performance with low computational complexity. All source code, documentation, and reproducible examples are publicly available.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.